QAOOSE 2006 Proceedings

10th ECOOP Workshop onQuantitative Approaches in Object-Oriented Software Engineering

3 July 2006 — Nantes, France

Edited by:

Michele Lanza, Fernando Brito e Abreu, Coral Calero, Yann-Gaël Guéhéneuc, Houari Sahraoui

Organizers

Fernando Brito e Abreu, Univ. of Lisbon, Portugal

Coral Calero, Univ. of Castilla, Spain

Yann-Gaël Guéhéneuc, Univ. of Montreal, Canada

Michele Lanza, Univ. of Lugano, Switzerland

Houari Sahraoui, Univ. of Montreal, Canada

Outline

QAOOSE 2006 is a direct continuation of nine successful workshops, held during previous editionsof ECOOP in Glasgow (2005), Oslo (2004), Darmstadt (2003), Malaga (2002), Budapest (2001),Cannes (2000), Lisbon (1999), Brussels (1998) and Aarhus (1995).

The QAOOSE series of workshops has attracted participants from both academia and industrythat are involved/interested in the application of quantitative methods in object-oriented softwareengineering research and practice. Quantitative approaches in the object-oriented field is a broadand active research area that develops andor evaluates methods, practical guidelines, techniques,and tools to improve the quality of software products and the efficiency and effectiveness of soft-ware processes. The workshop is open to other technologies related to object-oriented such ascomponent-based systems, web-based systems, and agent-based systems.

This workshop provides a forum to discuss the current state of the art and the practice in thefield of quantitative approaches in the fields related to object-orientation. A blend of researchersand practitioners from industry and academia is expected to share recent advances in the field-success or failure stories, lessons learnedand seek to identify new fundamental problems arising inthe field.

Contents

Metrics, Components, Aspects

“Measuring the Complexity of Aspect-Oriented Programs with Multiparadigm Metric”N. Pataki, A. Sipos, Z. Porkoláb . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

“On the Influence of Practitioners’ Expertise in Component Based Software Reviews”M. Goulão, F. Brito e Abreu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

“A substitution model for software components”B. George, R. Fleurquin, S. Sadou . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

Visualization, Evolution

“Towards Task-Oriented Modeling using UML”C. F. J. Lange, M. A. M. Wijns, M. R. V. Chaudron . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

“Animation Coherence in Representing Software Evolution”G. Langelier, H. A. Sahraoui, and P. Poulin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

“Computing Ripple Effect for Object Oriented Software”H. Bilal and S. Black . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

“Using Coupling Metrics for Change Impact Analysis in Object-Oriented Systems”M. K. Abdi, H. Lounis, and H. A. Sahraoui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

Quality Models, Metrics, Detection, Refactoring

“A maintainability analysis of the code produced by an EJBs automatic generator”I. Garćıa, M. Polo, M. Piattini . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

“Validation of a Standard- and Metric-Based Software Quality Model”R. Lincke and W. Löwe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

“A Proposal of a Probabilistic Framework for Web-Based Applications Quality”G. Malak, H. A. Sahraoui, L. Badri and M. Badri . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

“Investigating Refactoring Impact through a Wider View of Software”M. Lopez, N. Habra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

“Relative Thresholds: Case Study to Incorporate Metrics in the Detection of Bad Smells”Y. Crespo, C. López, and R. Marticorena . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

.

������������������������������������ �"!#�%$'&(���)��*+�-,-./����������012��%�%����3�546�7���8����� ���7�9�:;�%0������ ������;��* ?

=@?BADCE?GF�HJIKF IMLONQPSR�TSUMVXWYN[Z"=]\EP^IM_DT9`"=]abF UM_�H;=@?BL@=@acUMA

d�e fOg�hjilk:e mKiJnEoqp-hlnsrthGgEk:k:uvmwryx"gtmwrszwgErEe {�gtmO|~}�nEk:fwuvveBhl{�tils{xDnthGEmw|��m]uveBhl{juiQK]Ogt zwiQ�nto-Qm@ocnEhlk%g�iluv {

p-t k%EmK�pBileBh+{jBiGtmM�7s}lsE�y�zw|wgtfYe {i7wzwm]rgthj@stw7]E K¡EK ¢¤£¥�¦�§7£@s K¡EK -¢¨£¥�¦~©@§tªKs K¡EK -¢¨£¥

«�¬ ® ¯�°O±E®s²´³ {jfYe BijµnEhluve mKile7|¶f@hlnsrthGgEk:k:uvmwr¶· ³+¸ p�¹Oub{"gJf]hlnsk:uv{juvmwrJm]eBº{jnEo»ijº;gthle;|@e e vnEfwk:e mKiSile G¼wmwub½Kzwe-B[gtuvk:e7|¶iln+uvk:f@hlntse�BnM|]e�k:nM|]zwbg�hlub gtiluvnsm>gEmw|il¼]eBhleBocnthleJhle7|]z] eJ nsk:f]veB¾@uvi¿>nEo^nsÀMÁe Âij¿nthluve mKile7|yf]hlnErEhGgtk:{ Ã�n7º�e eBh7seB¾]gEBi½KzOgtmKiluviGg�iluvequbm]{jfYe Biluvnsm]{nsm�il¼]e�f@hlnsÀwve kÄ|]eÂiGgEuvv{-gthle�{iluvbKz]mO|@eBhDº;g KÃ7Qm�il¼wuv{fwgEfYeBhź;e¶|]e {jBhluvÀYeÆgykÇzwiluvfOg�hGgs|]uvrEk2{jnEo[iQºJg�hleÆk:eBijhluv�gEmw|�uil{ÅeB¾Mile mw{juvnEm�iln³ {jfYe BiGÈ]Ã]É9e�zw{je¶il¼]e¶k:eBijhluv¶iln´ nEk:fwz@ile¶{ijhlzwBilz@hGgE^ nsk:f]veB¾@uvi¿�nEo�gEv^il¼wenEÀ@ÁQe BijµnEhluve mKile7|Sgt{jfYe Bij¤hle bgtile7|ÇgEmw|>f@hlnK e7|]z@hGgE@ nsk:fYnEmwe mKil{DnEo ³+¸ p~ nM|]eEÃÉ�eile {ile7|%nsz@hJk:eBijhluv+nsm%iQº;n>ocz]mwBiluvnsmwgEv%e7½KzOgtYubk:f]ve k:e mKiGgtiluvnsm]{�ntoqÊ�n|@e {juvrsm%fOg�ijileBhlmw{�k%gs|@eJubm�gt{jfYe BijµnEhluve mKile7|yº;g7´gEmw|´ubm:f]z]hleÅnsÀMÁeBBijµnEhluve mKile7|{i¿]veÆgtmO|�Bnsk:fOg�hle7|�il¼]e�hle {jz]vil{ Ã

Ë ÌwÍ+ÎYÏSÐ�ÑÇÒ>Ó"ÎSÔlÐ+Í

ÕÖC�FB?ÂNc×t\ZSacIØÙI@_ÚN»VÙZD=]?GF IM_wF?B=@a»C>N»_�VÙ=YTYCt?B_\Â=MÛ¨FlܶIM?BCÆCt_^Ý@N»_SCtCE?ÂN»_SÝ^Þwß�Cs\lFBN[_^Ý^POASàSÝ@áSâOãN[_SÝy×�=Kä]Ct?�I@_%N[_D×�?BCEI@\ÂN»_SÝÆZ"Ct? ×�Ct_wF IMÝ@C+=MÛYFÂå^C\Â=MÛ¨FlܶIM?BCÅa[N[Û¤CE×tØY×�a»C@Þsæj_�\Â=MÛ¨FlܶIM?BCÅTYCE\ÂN[Ý]_:FÂå^CVç=w\lF+\ÂN»Ý@_SN[áD×tI@_wF;ZDIM?ÂF;=@ÛDFBåSCÆ×�=w\lFÅNc\Å\ÂZDCE_wFÅ=@_�FÂåSCÆV~I@N[_wFÂCE_^IM_D×�C¶=MÛ"FÂåSC�Z^?Â=YTYà^×�FEÞ@߶å^C×�=]\GF=@Û�\G=@Û¨FlÜ�I@?ÂC>VÙI@N[_wFBCt_^I@_^×�CÇåSN[Ý]åSa»ØÙ×t=@?B?ÂCEa»IMFÂCs\JÜ�NbFBå�FBåSCy\GFÂ?Bà^×�FÂàS? IMa"×�=@VÙZSa»C�âYN[FlØÙ=MÛFÂåSC�×t=YTYC@Þ^߶å^C%×�?BNbFBN»×EIMaqZ^I@?GF \�=MÛ�FÂå^C�\G=@Û¨FlÜ�I@?ÂC´×tI@_9ADC%NcTYCE_]FBNbáDCET9N»_FBåSC%CsIM?Ba[ØÚ\GFBI@Ý@Cs\=MÛ>FBåSCèTYCEä@Cta»=@Z"CtVÙCE_]F~ZS?B=Y×�Cs\Â\çÜ�N[FÂå'FÂåSCÖI@N»TÄ=MÛ:IéÝ]=O=OT'×�=]VÙZSa[CtâYNbFlØwãVçCsI@\ÂàS?BCtVÙCt_wFFÂ=O=@aQÞwê¶I]\GCsTÙ=@_�\Â=MÛ¨FlܶIM?BCÆVÙC�FÂ?BNc×t\ÅÜ�CÇ×EIM_~Ý]N[ä]CÆ?ÂCs×�=]VçVÙCE_^TSIKFBN[=]_^\+I@_^T~TYCtá^_SCÇ×t=YTYN[_^Ý×�=@_Oä]Ct_wFÂN»=@_^\�=]_FBåSC%TYCEä@Cta»=@Z^VçCE_wFÆ=MÛJ\G=]àS_^T-PSV~I@_^IMÝ]CEIMA^a[C´IM_^T�åOØOÝ@N»Ct_SNc×´×�=YTYC]ÞSëÅä@CE_FÂåS=]àSÝ@åìÝ@CE_SCt? IMaJ?BCE×t=@VÙVÙCt_^T^IKFÂN»=@_D\yÛ¤=@?�\ÂZ"CE×�N[Û¤ØON[_^Ýè\GCE_^\GN»ASa»C~VÙC�FÂ?BNc×t\:CtâONc\GFÚí î@ïKðP-FÂå^C×�=@_D×�?BC�FÂC´VÙCEI]\Gà^?ÂCEVçCE_wFFB=w=]a»\�I@?ÂCyFlØOZSNc×tI@a[a»Ø�Z^I@?BI]TYN[Ý]VçãBPYIM_DTÚacIM_SÝ]à^IMÝ]C�ãjTYCtZ"Ct_DTYCt_wFEÞ

æj_�FÂå^C´\Â=MÛ¨FlܶIM?BCÇTSCtä@CEa[=]ZSVÙCt_wF¶ZS?Â=Y×tCE\B\´ñ]ò7ó7ôQõÂñ@ötôQ÷µøKùSó´ZSacIØ�I�×tCt_wFÂ? IMa?B=@a»C@ÞYúÆ_9IMASã\lFB?BI]×7FÂN»=@_�Û¤=Y×�à^\ÂCE\Å=@_çFÂåSC�Cs\Â\ÂCt_D×�C�=MÛ-I´ZS?B=@ASa»CtVûIM_^TçC�âS×ta[à^TSCE\;FBåSC>\ÂZDCs×�NcIMaSTYCtFBIMN»ac\�í ü@ðQÞúÆA^\GFÂ? I@×�FÂN»=@_^\;TYCEZDCE_^Tç=@_çVÙI@_OØ:ÛµI@×�FÂ=@? \EýsàD\GCE?�?BCEþwàSN»?BCtVÙCt_wFB\EPsFÂCs× åS_SNc×tIMaYCE_OäwN»?B=@_SVÙCt_wFsPIM_^T�FBåSC�L]CtØ9TYCE\ÂN[Ý]_èTYCE×tN»\ÂN»=@_^\EÞ^æj_é\Â=MÛ¨FlܶIM?BC´FÂCs× åS_S=]a[=]Ý@ØIÙÿSñMõÂñ��M÷���� ?BCtZ^?ÂCs\GCE_]F \¶FÂå^CTYN[?BCE×�FÂN»ä@Cs\�N»_ ×�?BCEIMFÂN»_SÝ I@A^\GFÂ? I@×7FBN[=]_^\EÞ�߶åSC9Z^I@?BI]TYN»Ý@V Nc\çFÂåSCÖZS?BN»_^×�N»ZSa»C�AOØÜ�åSNc× å IZS?Â=]ASa»CtV�×tI@_ A"C�×t=@VÙZS?BCtåSCE_^TYCsT IM_DT TYCs×�=@VÙZ"=]\ÂCET N[_wFÂ= V~IM_^I@Ý@CsIMASa»C×t=@VÙZD=]_SCt_wF \í��tðQÞJæj_'ZS? I@×7FBN»×tC@PJI Z^IM? I@TSN[Ý]V TSN[?BCE×�FB\Ùà^\ÙN»_ÄNcTYCE_]FBNbÛ¤ØON»_SÝ FÂåSCÖCEa[CEVÙCt_wFB\çN»_'Ü�å^N»× å'IZS?Â=]ASa»CtV5Ü�N[a»aADC´TYCs×�=@VÙZ"=]\ÂCET�I@_^T~ZS?B=��lCs×7FBCET-ÞO߶åSCyZ^IM? I@TYN»Ý@V5\ÂC�FB\àSZÚFÂåSCy?Âà^a[Cs\I@_^T

? z]fwfYnEhjile |�ÀKçÊ� ¸ pq�@à �Kà �Kà ��������E�7��������s��Mà �

1

ZS?Â=]ZDCE?GFBN[Cs\tPOASàSF>IMac\G=Ù=��Ct? \FÂ=O=@ac\�Û¤=@?ÆTSCtä@CEa[=]ZSN»_SÝçI@ZSZSa»N»×EIKFBN[=]_^\tÞS߶å^CE\ÂCyVÙCtFÂåS=YTS\�I@_^TFÂ=O=@ac\�IM?BC´_S=MF�N»_^TYCtZ"Ct_DTYCt_wF>=@Û�FÂå^CtN»?ÆCt_OäON[?B=@_^VçCE_wF�N»_9Ü�åSN»× åFÂå^CtØÚ=Y×E×�àS?sÞ

߶åSC¶a»I]\lF����yØ]CEIM? \�=MÛ\G=@Û¨FlÜ�I@?ÂC¶TYCE\ÂN»Ý@_çå^I@\J\ÂCtCE_ç\ÂCtä]Ct? IMaOZS?B=@Ý@? IMVÙVÙN»_SÝ>Z^I@?BI]TYN»Ý@V~\Û¤?Â=]V(ñ��Yôjø��~ñKô����¶ÿõÂø �MõÂñ��!�ç÷¨ù"�ÙI@_^TçFÂå^C$#&%(';ß)'ÆúÇ< a»I@_SÝ@à^I@Ý@C¶N»_ÙFÂåSCÆVÙNcTOãQáSÛ¨FÂN»CE\EPMFB=ÿ"õÂøsö�� ���YõÂñ�*EÿõÂø+�@õÂñ��!�ç÷¨ù"�yÜ�N[FÂå%\GFÂ?Bà^×7FBàS?BCET´N»VçZ"Ct? IKFBN[ä]CJa»I@_SÝ@àDIMÝ@Cs\-,µú/.102%(.+PtHJI@\B×tI@a43�PFÂ=�FBåSCÖø@ò65��Bötô�7GøKõ7÷6�tù^ô�� �èZ^IM? I@TSN[Ý]VXÜ�N[FÂå)a»I@_SÝ@à^I@Ý@Cs\�a[N»L@CÙWOV~IMa»abF IMa»LP98):;:ûIM_DT=Æ=KÜ�Ctä@CE?EPNbF9Nc\N»VÙZD=]?GF IM_wF�FÂ= à^_^TYCt? \GFBIM_DT FÂåDIKF9_SCtÜ ZDIM? I@TYN»Ý@V~\Ú×EIM_S_^=MF9Ct_wFÂN»?ÂCEa[Ø?ÂCEZSa»I]×�C´FBåSC�ZS?BCtäON»=@à^\Æ=@_SCs\tPDASàYFÇ? IKFBåSCt?ÆÛ¤=@?BV(I~_^CtÜ \lFB?ÂàD×7FÂà^?BI@a�acIØ@Ct?�=]_9FÂåSC�FÂ=]ZÖ=MÛFÂåSCEV�Þ?%ÇA?�lCE×�FGã=@?BN[CE_wFBIKFBN[=]_�Nc\ÅI´_SCEÜ'Û¤=@?BV�=MÛC�âYZS?BCE\B\ÂN[_SÝ´?BCtacIKFBN[=]_^\JA"C�FlÜ�CtCt_~TSIMFBI:I@_^TÛ¤àS_^×7FBN[=]_^\EPså^=KÜCEä@Ct?sP�FBåSCE\ÂC�?ÂCEa»IMFÂN»=@_^\�N[VÙZSa»Nc×�N[FÂa»Ø%C�âYNc\lFBCET%N»_%FBåSC¶ZS?B=O×tCETYà^?BI@a@Z^I@?BI]TYN[Ý]V9Þ

߶åSC_^CtCET Û¤=@?Ù_SCEÜ ZS?B=@Ý]?BI@VÙVçN»_SÝ)Z^IM? I@TYN»Ý@V~\�N»\ÙI)?ÂCs\Gà^abF~=MÛÇFÂåSC9CEä@Ct?ÂãÝ@?B=KÜ�N[_SÝ×�=@VÙZSa»C�âYN[FlØ9=MÛ\Â=MÛ¨FlܶIM?BC@Þ9%ÇA?�lCE×�FGã=@?BN[CE_]FBCET9ZS?B=@Ý]?BI@VÙVçN»_SÝ@,�%;%ÇHA3>N»\ÇÜ�N»TSCta»ØÖà^\GCsTèN»_FÂåSC¶\G=@Û¨FlÜ�I@?ÂC+N[_^TSà^\lFB?ÂØyÛ¤=]?�V~IM_DIMÝ@N»_SÝÆacIM?BÝ@C+ZS?Â=��lCE×�FB\EPEASàSF�?BCE×tCt_wFÂa»Ø:\G=]VçC=MÛYFBåSCÜ�CEI@Lwã_SCE\B\GCs\ÅCEVçCE?ÂÝ]CET-Þ]HÅ?B=@ASa»CtV~\a[N»L@C>×t?Â=w\Â\Gãj×�àYFÂFÂN»_SÝ�×�=]_^×�CE?Â_D\tP]V�àSa[FÂN[ãTSN[VÙCt_D\GN»=@_^I@aD\ÂCtZ^I@?BIMãFÂN»=@_�=@Û^×�=]_^×�CE?Â_^\EP\GØOVÙVÙC�FÂ?BNc×ÅC�âOFBCt_^\ÂN[=]_%=MÛ"I>×�acI@\B\�åSN[CE?BI@?B× åOØçí î]îðYIM?BCÅå^I@?BT´FB=ÇåDIM_^TYa»C@ÞÕÖ=YTYCt?B_9ZS?Â=]Ý@? IMVÙVÙN[_^Ý�acIM_SÝ]à^IMÝ]CE\¶å^Iä]C´VÙI]TYC:Z"=]\B\GN»ASa»C´FÂåSC%A^N[?ÂFÂåÖ=@Û�_SCEÜ Z^?Â=]Ý@? IMVçãVçN»_SÝ:Z^I@?BI]TYN»Ý@V~\;a[N»L@CCB�D�EFEHG´ô����ÆÿI*[ñKô��(�J�tôjñBÿ"õBø+�@õÂñ��K�ç÷¨ù"�L,µß�ÕÖH�3�íM�ðPN�N�tùO�tõ7÷µö+ÿ"õBø�7�MõÂñ��!�ç÷¨ù"�P,0yH�3´í îMüMðPSIM_^TñKó¿ÿQ� ö�ô�7GøKõ7÷6�tù^ô����yÿõÂø+�@õÂñ��!�ç÷¨ù"�R,µú$%ÇH�3yí��S�ðÞ

WO=MÛ¨FlܶIM?BCÇVÙCtFÂ?BN»×E\�å^Iä@C´IMa»Ü¶IØO\�A"CtCE_è\lFB?Â=]_SÝ@a»Ø~?ÂCEa»IMFÂCsT�FB=ÙFÂåSC:Z^I@?BI]TYN»Ý@V6à^\ÂCET�N»_FÂåSCÙ?BCE\ÂZDCs×7FÂN»ä@CçZ"Ct?BN[=YT-Þ-߶åSC~Õè×S8�IMA"CRDHT@ö�*[ø��~ñKôQ÷µö�öBø��ÆÿI*�� U@÷¨ô�TÖ_OàSV�A"Ct?ÙíîðÅܶI@\yTYC�ã\GN»Ý@_SCsT~Û¤=]?�VÙCsI@\ÂàS?ÂN»_SÝ�FÂåSCyFBCE\GFÂN»_SÝ�C��"=]?GF \¶=MÛ�_^=@_Yãj\lFB?ÂàD×7FÂà^?BI@aO#H%('�ß)'>ú>< ZS?B=@Ý@? IMV~\EÞHÅN[Ü�=KܶIM?BLO\ÂNSíV�SWðQP�>Æ=KܶIKFGF;IM_^T:ê¶I@L@Ct?í��N��ðwC�âOFÂCE_^TYCET:FÂåSC¶×�ØY×�a»=@V~IMFÂNc×Å×�=]VçZ^a[CtâON[FlØ´Ü�NbFBåFÂåSC%_^=MFÂN»=@_Ö=@Û�ùO��ó ôQ÷¨ù"�X*���Y��Z*;N»_è=@? TYCt?¶FB=ÚTYCs\Â×t?ÂN»ADC�\lFB?Âà^×�FÂàS?BCET9Z^?Â=]Ý@? IMV~\�ADCtFGFÂCE?EÞú�Û¨ãFÂCt?çFBåSC=]A?�lCE×�FGã=@?BN[CE_wFÂCET Z^IM? I@TYN»Ý@V A"CE×tI@VÙCÜ�NcTYCta»Ø I@×E×�CEZYFÂCsT IM_^T à^\ÂCET-P;A"=MFÂåÄFÂå^CI@×tI]TYCtVÙNc×�Ü=]?ÂacT-P@I@_^T�FÂåSC�ælß'N»_^TYà^\GFÂ?BØ%Û¤=Y×�àD\GCsT�=]_ÙVçCtFÂ?BN»×E\JA^I]\GCsT�=]_~\GZ"CE×tN»I@aS=@A?�lCs×7FÂã=@?BN[CE_]FBCETìÛ¤CsIKFBàS?ÂCs\tP;a»N[L]C)ù[�?�Ùò��tõ�ø]\�ö�*[ñó ó���ó P(���jÿ"ô_^ ø`\�÷¨ùa^Q��õ7÷¨ôjñKùö �Öô¿õ+���tP¶ùQ�?�~ò �tõ�ø]\ö ^Y÷b*��Kõ+�tù'ö�*[ñKóBó���ó7PqC�F ×MÞJí ïKðÞ�WOCtä]Ct? IMa�N[VÙZSa»CtVÙCt_wF IKFÂN»=@_D\´=MÛ\Âà^× å VçCtFÂ?BN»×E\´IM?BCçIäKIMN»acIMASa»CÛ¤=@?-FBåSCÅVÙ=]\GFqZD=]ZSàSacIM?qacIM_^Ý@à^I@Ý@CE\c,µa[N»L@C-

>ÆCt?BC�Ü�C�Ý@N»ä@CÙFBåSC\GFÂ?Bà^×�FÂàS?BC�=MÛ�FBåSC�Z^I@ZDCE?EÞ�æj_ \ÂCE×7FBN[=]_ î9Ü�CÚTYC�áD_SCÚ=]àS?�V�àSabFBNbãZ^IM? I@TYN»Ý@V�VçCtFÂ?BN»×@PKFÂå^C/i2j176�MõBñBÿf^YÞ@ú�Û¨FBCt?JFBå^IKFÜC�TYCtá^_SC�FBåSCÆ×�=]VçZ^a[CtâON[FlØ�=@ÛDFBåSCÆ×�acI@\B\EPÜ�åSCt?BCy×ta»I]\Â\Nc\�TSC�á^_SCsTÚI]\�I�\GCtF�=MÛ�TSIMFBIX,µIKFÂFÂ?BN[A^àYFÂCs\ 3�IM_DTÚ×�=]_wFÂ?B=@aD\GFÂ?Bà^×�FÂàS?BCE\F,¤VÙCEV�ãADCE?>Û¤àS_D×7FÂN»=@_D\tPVÙC�FBåS=YTS\ 3>×tI@?Â?BØON[_SÝÚ=@àYFy=@Z"Ct? IKFÂN»=@_D\�=@_ÖFBåSCÙIKFGFB?ÂN»ASàYFBCE\EÞ"úÆÛ¨FÂCE?ÂܶIM? TS\EPN[_è\ÂCE×�FÂN»=@_Öü~Ü�C:C�âYZSacIMN»_ÖåS=KÜ =]àS?ÆVÙC�FB?ÂNc×%I@ZSZSa»N[Cs\¶FÂ=Úú$%ÇH _^=MFÂN»=@_D\ÆIM_^TÖ×t=@_^\GFÂ?Bà^×�FB\EÞWOCE×�FÂN»=@_R�TYCs\Â×t?ÂN»ADCs\>=]àS?yFÂCs\lF´?ÂCs\Gà^abF \ÇÜ�å^Ct_)I@ZSZSa»ØON[_SÝÚFÂåSCÙVÙC�FB?ÂNc×�FÂ=FÂåSCk%;%ÇH I@_^Tú$%ÇH ä]Ct? \GN»=@_D\�=@ÛDFBåSC$0y=N#)TYCE\ÂN»Ý@_ÙZ^IKFÂFÂCE?Â_^\EÞ�l CÆC�âYZSacIMN»_�å^=KÜ'ú$%ÇHI��CE×�FB\�FBåSC>×�=]VçãZSa[CtâYNbFlØ)=MÛ¶FÂåSCs\GC~N»VÙZSa»CtVÙCt_wFBIMFÂN»=@_^\EP�I@_^T N[_Ü�å^N»× å×tI]\GCs\:ú$%ÇH ZS?B=KäwNcTYCsT)IÖ\ÂN[VÙZSa»Ct?\G=]a[àYFBN[=]_qÞ

m npo Ò$qlÎYÔ]r$s;ÏfsJÑ>Ô]t�o ovu-ÎYÏ^ÔGÓ

߶åSC)Ü�Cta»a[ãQLO_S=KÜ�_ VÙCEI]\Gà^?ÂC)=MÛ�Õè×S8�IMA"C@PFÂåSCì×�ØY×ta[=]VÙIMFÂNc×é×t=@VÙZSa»C�âYNbFlØ2íîð:Nc\ADI@\ÂCET=@_Sa»ØÚ=]_FBåSC%_OàSV�ADCE?Æ=MÛ;ZS?BCETYNc×tIMFÂCE\¶N»_èIÙZS?B=@Ý]?BI@V�ý

V (G) = p+ 1Þ^߶åSC:N»_^I@TSCEþwà^I@×tØ

=MÛ+FBåSCçVçCsI@\ÂàS?BC�A"CE×t=@VÙCE\´×�a»CEI@?EPN[ÛÜ�C�?BCEI@a[NVhtC�FÂå^IMF%×�ØY×�a»=@V~IKFBN»×ç×�=]VÙZSa[CtâYNbFlØÖN»Ý@_S=]?ÂCs\FÂåSC:_SCs\lFBN[_^Ý~a[CEä@Cta-=@Û�FBåSC:ZS?BCETYNc×tIMFÂC´_S=YTYCs\tÞYæjVÙZS?B=Kä@CEVçCE_wFB\¶I@\�Ü�CtN»Ý@åwFBN[_SÝ�FÂå^C%×�=@_wFB?Â=]a\lFB?Âà^×�FÂàS?BCçÜ�NbFBåèFÂå^Cç_^CE\GFÂN»_SÝa»Ctä]Cta�ÜCE?ÂC�ZS?B=@Z"=]\ÂCETÖAOØ=>>I@?Â?BNc\G=]_éIM_^T)ÕèIMÝ]Cta�í����KðPDAOØHÅN[Ü�=KܶIM? \GLON+íV�SWðÅIM_^T)AOØ=>Æ=KܶIKFGF:IM_DTéê¶IML]Ct?çí��N��ðÞ߶å^Cótö øBÿQ�Ù=@Û�IZS?ÂCsTYNc×tIKFBC�_^=OTSCN»\~I \GCtF~=MÛy\GFBIMFÂCEVçCE_wFB\EPJÜ�åS=]\ÂCCtâYCE×�àSFÂN»=@_ÄTSCtZ"Ct_^TS\�=@_ FBåSCÖTYCE×tN»\ÂN»=@_'VÙI]TYC�N[_'FÂå^CZS?ÂCsTYNc×tIKFBC_^=OTSC@ÞÅ߶åSC9_SCs\lFBN[_SÝ a»Ctä@CEa�=MÛÇIì\GFBIMFÂCtVÙCE_]F~Nc\ÙTYC�á^_^CET'I@\�FÂåSCÖ_OàSV�ADCE?Ù=MÛZS?ÂCsTYNc×tIKFBC´_S=YTYCE\¶Ü�åS=w\GC:\B×�=]ZDC:×�=]_wFBIMN»_^\�FBåSC%\lF IKFBCtVÙCt_wFEÞ

߶åSC´_SCE\GFÂN»_SÝÙa[CEä@CEa-_S=MF IKFÂN»=@_�?BCZw^Cs×7FB\�ZS?B=Y×�CsTYàS? IMaZS?B=@Ý@? IMV~\N[_ÖI@_�I]TYCEþwà^IMFÂCyÜ�IØ]Þú�FÅFÂåSCÆ\BIMVÙC¶FBN[VÙC@P]NbF�TY=OCE\J_S=@FJFBI@L@C�N»_]FB=%I@×E×�=@à^_]FÅTSIMFBI´å^I@_^TYa»N[_^Ý^PMÜ�åSNc× åÙå^I@\+×�Ct_wFB?BI@a?Â=]a[C:N»_ÖVÙ=OTSCt?B_ÖZS?Â=]Ý@? IMVÙVÙN[_^ÝçacIM_^Ý@à^I@Ý@CE\EÞ?l C%å^Iä@C´FB=~FBI@L@C´FBåSC�×�=]VÙZSa[CtâYNbFlØ�=MÛ;FÂå^CTSIKF I�TYC�á^_^CET:IM_DT>FBåSC+×t=@VÙZSa»C�âYNbFlØÇ=MÛYTSIKF I¶å^IM_DTYa[N»_SÝ�N»_wFÂ=>×t=@_^\ÂN»TSCt? IKFÂN»=@_�Þ ú�×E×�=]?BTYN»_SÝ]a[Ø]PFÂåSC!ixjÄö ø��>ÿO*V�+U]÷¨ô�T~=MÛJI�Z^?Â=]Ý@? IMV/Nc\�Iç\ÂàSVX=MÛ�FÂåS?BCtC:×�=]VÙZD=]_SCt_wFB\Eý�@ÞJDÅøKù^ôQõÂø�*Jó7ô¿õy�^ötô�Yõ ��ø`\ÇÿõÂø+�@õÂñ��Jz�ÕÖ=]\GF%ZS?B=@Ý]?BI@VÙ\Çå^Iä]C�FBåSC�\ÂI@VÙCÙ×t=@_wFÂ?B=@aÅ\GFBIMFÂCtã

VÙCt_wFB\�N[?B?BCE\ÂZDCs×7FÂN»ä@CEa[Ø)=MÛ¶FÂå^CÚZ^I@?BI]TYN[Ý]V à^\ÂCETqÞ�߶åSC�×�=@_wFB?Â=]a+\lFB?ÂàD×7FÂà^?ÂC�Nc\%?BCtZ^?ÂCtã\GCE_wFÂCET�AOØ%IyÝ@? IMZSå�Ü�åSCt?BC_^=OTSCE\JIM?BC¶\lF IKFÂCEVÙCt_wFB\;IM_DTX,¿TYN[?BCE×�FÂCsTQ3�CETYÝ]CE\�?ÂCEZS?BCE\ÂCt_wFFÂåSC¶Z"=]\B\GN»ASa»C-w^=KÜ =MÛD×t=@_wFÂ?B=@aQÞs=YTYCE\�Ü�N[FÂåçVç=]?ÂC+FÂå^I@_�=]_SC=]àYFÂZ^àYFJCETYÝ]CIM?BC×EIMa»a[CsTZS?BCETYNc×tIMFÂC´_S=YTYCE\EÞSÿO*�� U]÷¨ôTÚø]\2�@ñKôjñ~ô�TBÿQ�7ó�zYæF�?BCZw^Cs×7F \+FBåSC:×�=]VçZ^a[CtâON[FlØ~=MÛ;TSIKF I�àD\GCsT@,¤a»N»L@C´N[_FÂå^C×tI]\GCÇ=MÛ�×ta»I]\Â\ÂCE\�37Þ"{ÇIMFBI�_S=YTYCE\�IM?BC>?BCtZS?BCE\ÂCt_wFBCET~I@\�TYN��"CE?ÂCE_wFFlØOZ"CÇ=@Ûq_S=YTYCE\�N»_�FÂå^C×�=]_]FB?Â=]aÝ@? IMZSåqÞ

ïSÞJDÅø��>ÿO*�� U]÷¨ôTéø]\k�@ñKôjñéñ]ö ö ��ó ó�z&8=]_S_SCE×�FÂN»=@_)ADCtFlÜCECt_ì×�=@_wFB?Â=]a;\GFÂ?Bà^×�FÂàS?BC~IM_^T TSIKF IN»\�?ÂCEZS?ÂCs\GCE_wFÂCETÄAOØ CsTYÝ@Cs\~ADCtFlÜCECt_'FÂå^CéTSIKF Iì_S=YTYCE\ÚIM_^T \GFBIKFBCtVÙCt_wF \ÙFÂå^IMF�à^\ÂCFÂåSCEV9ÞS߶åSCÇTYN»?BCE×7FBN[=]_Ú=@Û-FÂåSCs\GCyCETYÝ]CE\?ÂC�w^CE×�FFBåSC´TSIKF I;w^=KÜ|,¤Û¤=@?¶C�âSIMVÙZSa»C>N»_×EI@\ÂC=MÛ+?BCEI]TYN»_SÝÚTSIMFBIÚFÂåSCçCETYÝ]C�Nc\´TYN[?BCE×�FÂCsT9Û¤?Â=]V FÂåSCÙTSIMFBI�_^=OTSC�FB=ÚFBåSCç\GFBIMFÂCEVçCE_wF�37Þ{ÇIMFBIÙCsTYÝ@Cs\�I@?ÂC´_SCs\lFBCET�AOØ~FÂå^C:_SCE\GFÂN»_SÝÙa[CEä@CEa-=MÛ�FÂåSCEN[?Æ\GFBIMFÂCtVÙCE_]FÆ_^=OTSC@Þ

úÆ_èN»VÙZD=]?GF IM_wF>Û¤CsIKFÂà^?ÂC%=@ÛÅ=@àS?ÇVÙC�FÂ?BNc×%Nc\ÆFBå^IKFyNbF:TY=OCE\>_S=MF´×�=@à^_]FÇFÂåSCç×�=]VçZ^a[CtâON[FlØ�=MÛTSIKF I:å^IM_DTYa[N»_SÝ�A^I@\ÂCETÙ=]_ÙFÂåSCÇZSa»I]×�CÇ=MÛFÂåSCyTYCE×ta»I@?BIMFÂN»=@_qÞw߶åSC>VÙCtFÂ?BN»×�CE_^×�=]àS_wFÂCE?B\ÅFÂå^IMFäI@a[à^CÇCtâYI]×7FBa[Ø~IKF�FÂå^CyZ"=@N»_wF�=@Û�TSIMFBI�å^IM_^TSa[N»_SÝ^ÞQ%>Û�×t=@àS? \ÂC@P]FÂåSCyVÙC�FÂ?BNc×ÇI@a»\Â=�VÙCEI]\Gà^?ÂCs\FÂåSC9TYCE×ta»I@?BIMFÂN»=@_N»_'I@_ N»VçZ^a[Nc×�N[FÂC�Ü�IØý�a[=Y×EIMa�äI@?ÂNcIMA^a[Cs\�I@?ÂC�àD\GCsT =@_^a[ØìN»_ FÂå^C�a»=Y×tI@a×�=YTYC:×�=]_]FBC�âOFx,¤N»_�FÂåSC%\ÂàSASZ^?Â=]Ý@? IMVk37Þ

3

߶åSCt?BC%Nc\yI@_è=MFBåSCt?yZ"=]\B\GN»ASa»C�ܶIØÚFB=ÚÝ@CtF>FBåSCE\ÂC�?BCE\ÂàSa[FB\EÞO.�C�Fyà^\´\GàSZ^ZD=w\GC�FÂå^IMFÇÜ�Cå^Iä@C9_S= TSIMFBI _S=YTYCs\�IM_^T TSIMFBIìCsTYÝ@Cs\ÙN[_ =]àS?~Ý@? IMZSå�PJASàYFÚÜ�CÖ?ÂCEZSacI@×�C�FBåSCtV�Ü�åSN[F\GZ"CE×tN»I@a×�=]_wFÂ?B=@aÅ_S=YTYCs\týe}?BCEI]TYCt?]~ÙI@_^T[K=@?F}¨Ü�?ÂN[FÂCE?`~7Þ�߶åSCs\GC×�=@_wFB?Â=]aÅ_S=YTYCE\�=@_Sa»Øì\GCE_^TIM_^T'?BCE×�CEN[ä]C9N[_YÛ¤=]?ÂV~IMFÂN»=@_qÞ+߶å^CtØ Ü�N»a»a>A"CÖN[_^\ÂCt?ÂFÂCsTP�là^\lFADCtÛ¤=@?BCèIM_^T IKÛ¨FBCt?~FBåSCè?BCEI@a×�=@_wFB?Â=]a"_^=OTSCE\¶Ü�åSNc× å9?ÂCsI@TÚI@_^T[K=@?¶Ü�?ÂN[FÂC:TSIMFBI^ÞS߶åSC´_SCE\GFÂN»_SÝ~TYCtZSFÂåÖIM_^T9×t=@VÙZSa»C�âYNbFlØäI@a[à^CÆÜC>Ý@C�FÜ�NbFBåÙFÂåSNc\+VÙ=YTYCEa^Nc\JFÂå^C/�@\BIMVÙCÆI]\;FÂåSCÇ=@_SCÇ×tIMac×�à^a»IMFÂCETçÜ�N[FÂåÚú1 Ý]?BI@ZSå^\EÞ

l)C>×EIM_Ù_^IMFÂàS? IMa»a[Ø�C�âOFÂCE_^Tç=@àS?+Vç=YTYCEaSFB=%=@A?�lCs×7FÂãQ=]?ÂN»Ct_wFÂCsT�Z^?Â=]Ý@? IMV~\tÞK߶åSCÇ×�Ct_wFB?BI@a_S=MFBN[=]_é=MÛJFBåSCÙ=@A?�lCs×7FÂãQ=]?ÂN»Ct_wFÂCsT9Z^IM? I@TSN[Ý]VXNc\ÇFÂåSCÙ×�acI@\B\tÞ߶åSCE?ÂCtÛ¤=@?BC�Ü�C�TYCs\Â×t?ÂN»ADCçåS=KÜÜC´VÙCEI]\GàS?BCyFÂåSC%×t=@VÙZSa»C�âYN[FlØ�=MÛJIÙ×�acI@\B\�á^?B\GFEÞI%Ç_ÚFBåSC:A^I@\ÂC´=MÛ�FÂåSC:Z^?ÂCEäwN»=@àD\�\GCs×7FÂN»=@_D\ÜCÖ×tI@_'\ÂCtC9FBåSCè×�acI@\B\ÙTYC�áD_SNbFBN[=]_ I@\~Iì\GCtF�=MÛJ,¤a»=O×EIMab3ÙTSIKF IìIM_^T'I\GCtF~=MÛ´VçCtFÂåS=YTS\I@×t×tCE\B\GN»_SÝ�FÂåSCEV9Þ

߶åSC×t=@VÙZSa»C�âYNbFlØ´=@ÛDI>×ta»I]\Â\�N»\�FÂå^C�\ÂàSV =MÛ^FÂåSC¶×�=]VçZ^a[CtâON[FlØ´=MÛYFBåSCVÙCtFÂåS=YTS\;IM_DT:FÂå^CTSIKF I´VçCEV�A"Ct? \g,µIMFGFB?ÂN»ASàYFBCE\�37Þ]úÇ\JFBåSCÆ×�=]_]FB?Â=]aS_S=YTYCs\g,¤_S=YTYCs\JA"Cta»=@_SÝ]N[_SÝ´FB=:FÂå^CÆ×�=@_wFB?Â=]a\lFB?Âà^×�FÂàS?BC:=MÛ;=@_^C%=MÛ�FÂåSC�VçCEV�A"Ct?ÆÝ]?BI@ZSå^\�3�ÜCE?ÂC´àS_^N»þwàSC]PSFÂå^Ct?BC:N»\�_^=�Z^IKFBå�Û¤?Â=]V =@_^CVçCEV�A"Ct?�Ý]?BI@ZSåÚFB=~IM_S=@FÂåSCE?�=@_SC]Þf>Æ=KÜ�Ctä@CE?EPwFBåSCt?BC:×�=@à^a»TADC%IMFGFB?ÂN»ASàYFBCE\2,¿TSIKF I�_^=OTSCE\�3Ü�åSN»× åÄIM?BCÚà^\ÂCETAOØìVÙ=@?BC�FBå^IM_ =]_SCÚVÙCEV�A"Ct?ÙÝ@? IMZSå�Þ�߶åSCs\GC�IKFGFB?ÂN»ASàYFBCE\�å^Iä@C�TSIKF I?ÂCtÛ¤Ct?BCt_^×tCyCsTYÝ@Cs\FÂ=�TYN��"CE?ÂCE_wFÆVçCEV�A"Ct?�Ý]?BI@ZSå^\EÞ

߶åSN»\�N»\�I _^IKFBàS?BI@a�VÙ=OTSCta>=@Û´FÂåSCè×ta»I]\Â\EÞÅæFÚ?BCZw^Cs×7F \ÙFÂåSCèÛµI@×7F~FBå^IKF�I×�acI@\B\~N»\ÚI×�=@å^Ct?BCt_wFç\ÂC�F~=MÛyIKFGFB?ÂN»ASàYFBCE\,µT^IKFBI"3�I@_^T FÂåSC�VÙC�FÂå^=OT^\ÙÜ=]?ÂLON»_SÝè=@_ FÂåSCÖIMFGFÂ?BN»ASàYFÂCs\tÞ>ÆCt?BCyFÂåSC�J�tôb^^ø��Kók,¤VÙCtV�ADCE?�Û¤àS_^×�FÂN»=@_^\�3¶IM?BC´ZS?Â=Y×tCETYàS?BCE\�?BCtZS?BCE\ÂCt_wFBCETAwØÚN[_^TSN[äONcTYà^IMaúÄÝ@? IMZSå^\c,¤FÂåSCVÙCtV�A"Ct?�Ý@? IMZSåD\ 3�Þ�ëÅä@CE?ÂØyVÙCtV�ADCE?�Ý@? IMZ^åyåDI@\-N[FB\�=KÜ�_:\lF IM?ÂF�_S=YTYC+I@_^TFÂCt?BVÙN[_DIMa�_^=OTSC@PDI]\¶FÂå^CtØ9IM?BC:N[_^TSN[äONcTYà^IMa»a»Ø�×EIMa»a»I@ASa[C´Û¤à^_^×7FBN[=]_^\tÞOl å^IMF>V~IML@Cs\¶FÂå^N»\Ç\GCtF=MÛ^ZS?B=Y×�CsTYàS?BCE\�Vç=]?ÂCFÂåDIM_�I@_�=]?BTSN[_^I@?ÂØya»N[AS? IM?BØ´N»\�FBåSC�×�=@VÙVÙ=@_ç\GCtF�=@Û"IMFGFÂ?BN»ASàYFÂCs\�à^\GCsTAwØ9FBåSCÙVÙCtV�A"Ct?:Z^?Â=Y×�CsTYàS?BCE\EÞO>>Ct?BC�FBåSC~IKFÂFÂ?BN[ASàSFÂCE\´I@?ÂCç_S=MF:a»=Y×tIMa�FB==]_SC�Z^?Â=Y×�CsTYàS?BCASàYFÆa»=O×EIMaFB=ÙFÂåSC:=]A?�lCE×�FEPSIM_DT×EIM_9ADC:I]×t×�Cs\Â\ÂCETAOØÚ\ÂCtä@CE?BI@aZS?Â=Y×tCETYàS?BCE\EÞ

.qC�F¶à^\�×�=]_^\GNcTYCE?+FBå^IKF�FÂåSC´TYCtá^_SN[FÂN»=@_Ú=@ÛqFÂå^Cyú12Ý@? IMZSå~Z"Ct?BVÙNbF \+FBåSCyCtVÙZYFlØ�\ÂC�F¶=MÛ×�=@_wFB?Â=]a-_S=YTYCE\EÞSæj_9FÂåDIKFÇ×EI@\ÂCyÜ�C:Ý@CtF>I~×�acI@\B\GNc×tI@aqTSIKF I~\lFB?ÂàD×7FÂà^?ÂC]Þ^߶åSC%×�=]VÙZSa[CtâYNbFlØ=MÛI×ta»I]\Â\ÂN»×EIMa�TSIKF I\GFÂ?Bà^×7FBàS?BC�Nc\>FBåSCÙ\GàSV =@ÛÅFÂå^CçTSIMFBI_S=YTYCE\EÞ-߶åSCç=@ZSZ"=]\ÂNbFBC�\ÂNbFBà^IKFBN[=]_N»\�IMac\G=�ZD=w\Â\ÂN[A^a[C]ÞNl åSCE_ÚIF}l×�acI@\B\`~�×t=@_wFBI@N[_^\TYNc\_�làS_D×7F¶VçCtFÂåS=YTS\-�FBåSCt?BC>Nc\_S=�×t=@VÙVÙ=@_TSIKF I%\GåDIM?BCETçADCtFlÜCECt_~FBåSCtVP@Ü�C>×�=]VçZ^àYFÂC�FBåSCÇ×t=@VÙZSa»C�âYN[FlØ�=@Û"FBåSCÇ×ta»I]\Â\+I@\JFBåSCÇ\ÂàSV=MÛÅFBåSCç×t=@VÙZSa»C�âYN[FÂN»CE\Ç=MÛÅFBåSCçTSN»\6�làS_^×�FyÛ¤à^_^×7FBN[=]_^\tÞIl Cç×EIM_èNcTYCE_]FBNbÛ¤Ø9FBåSNc\:×�=@_D\lFB?Âà^×�FyI]\IM_�=@? TYN[_DIØÙÛ¤àS_^×7FBN[=]_9a[N»AS? IM?BØ@Þ

߶åSCE\ÂCÖC�âSIMVÙZSa»CE\ÚIMac\G=ìZ"=@N»_]FÚFÂ=FBåSCÖÛµI@×�F~FÂå^IMFÚÜ�Cèà^\ÂCÖZ^IM? I@TYN»Ý@VçãN[_DTYCtZ"Ct_^TSCt_wF_S=MFBN[=]_^\tPJ\Â=éÜ�C�×EIM_'IMZSZSa»Ø=@à^?çVÙCsI@\ÂàS?ÂCÚFÂ= ZS?B=O×tCETYà^?BI@a¿P�=@A?�lCs×7FÂãQ=]?ÂN»Ct_wFÂCsT-P�=@?ÙCtä]Ct_VçN[âYCETOãj\GFlØwa»C´ZS?B=@Ý@? IMV~\EÞw߶å^N»\�ܶI@\�=]àS?�Ý@=wIMaQÞ

n"r;u�Ó"Î"� Ï^Ô]u�ÍÎ?u�ÑéÏSÐ-t+Ïfscovo5ÔÂÍgt

ú>\ÂZDCs×7FGã=@?BN»Ct_wFÂCsTÙZS?Â=]Ý@? IMVÙVÙN[_^ÝJ,¿ú$%ÇHA3JNc\=@_SCÆ=@Û-FÂåSC>Vç=w\lF�ZS?B=@VÙN»\ÂN»_SÝ:_SCtÜ \Â=MÛ¨FlܶIM?BCTYCtä]Cta»=@ZSVÙCt_wF�FÂCE× å^_SN»þwàSCs\tÞ;ú$%ÇHûI@N»T^\�IéADCtFGFÂCE?Ùå^IM_^TSa[N»_SÝé=@ÛçötõÂøó ótöZ�YôQôQ÷¨ù"�ìö øKù"ö��tõ7ùSóí��îðP]I@\×�=]VçZDIM?BCET�FÂ=�=@A?�lCs×7FGã=@?BN»Ct_wFBIMFÂN»=@_qÞ@߶åOà^\ú$%ÇH Nc\+I:Ý]Ct_SCE?BIMFÂN»ä@C�ZS?B=@Ý]?BI@VçVÙN»_SÝZ^IM? I@TYN»Ý@V FBå^IKF�I@N[V~\yFB=�å^Cta»Z N»_ Ü�?ÂN[FÂN»_SÝ9VÙ=@?BCçVÙ=YTYàSacIM?BNVhtCETqPqIM_^T)VÙ=@?BCçV~I@N[_wFBI@N[_SãIMASa»C~×�=YTYC@Þqß�=YTSIØI \´ú$%ÇH2N»VçZ^a[CEVçCE_wFBIKFBN[=]_^\C,µI@VÙ=@_SÝ�Ü�åSNc× å)FBåSCÙVÙ=]\GF%Ü�N»TYCEa[Øwãà^\ÂCETN»\yú>\ÂZDCs×7F�

àSa»I@?ÂNVhtN»_SÝ�×t?Â=w\Â\B×�àYFÂFÂN»_SÝÚ×t=@_^×tCt?B_^\�N»_]FB=�=@Af�lCE×7FÂãQ=]?ÂN»Ct_wFBCET�ZS?B=@Ý@? IMV~\EÞ^ú$%ÇH TYCtá^_SCE\>FÂå^CÛ¤=@a»a[=KÜ�N»_SÝçN[VÙZ"=@?ÂFBIM_wF>×t=@_^\GFÂ?Bà^×7F \tPYI]\GNcTYCyFÂåSC!%;%ÇH _S=@FÂN»=@_^\Eý

�@Þ$�øK÷¨ùDôö��Yôg���ÅùD÷¨ô¿÷µøMùSó�I@?ÂC:V~I@TSC%àSZÖ=MÛJH;=@N»_wFB×tàYFÆFlØOZDC]P"I@_^T9H;=@N»_wFB×�àSFÇ\GN»Ý@_DIKFÂà^?ÂC]Þ߶åSCèZ"=@N»_wFB×�àSF~FlØwZ"CéTYCs\Â×t?ÂN»A"CE\1^^ñKô~å^I@ZSZDCE_^\EPÅC@Þ Ý^Þ)�?aì\GFBIM_DTS\ÙÛ¤=@?�Û¤àS_^×7FBN[=]_×tI@a[aQP-�fQS?O��)\GFBI@_^TS\�Û¤=@?çFÂåSC�C�âYCs×�àYFBN[=]_ =MÛyIÖÛ¤àS_^×�FÂN»=@_qÞÅ߶å^C�\ÂN[Ý]_^IKFBàS?BCÚTYC�ã\Â×t?ÂN»A"CE\1^O÷µö�^ LwN»_^T'=MÛyÛ¤àS_^×�FÂN»=@_^\ÚIM?BC�VÙ=]_SNbFB=@?BCETÄAOØ FÂå^CÖZD=]N[_wFB×tàYFTYCtá^_SN[FÂN»=@_qÞ+ [� a�?¢¡ �£?¤"¤ VçCsIM_^\�IMa»aÅFBåSC~Û¤àS_^×�FÂN»=@_^\�Ü�N[FÂåFBåSCÚ_^I@VÙC¡^P�FÂåDIKF�å^Iä]CCtN[FÂåSCE?�I [��=]?�I@_@�?ç?BC�FBàS?B_~FlØOZDC]PSIM_^TÚ?ÂCs×�CtN»ä@C>=@_SCyZ^IM? IMVÙCtFÂCt?�=MÛ�IM_OØçFlØwZ"C@Þ

îYÞ$�øK÷¨ùDôö��Yô¿ó�IM?BC%\GCtFB\Æ=@ÛJZD=]N[_wFB×tàYFyTSC�á^_SN[FÂN»=@_D\ÆA"=@àS_^T9AOØ�H;=@N»_]F ×�àYFÇ=@Z"Ct? IKFÂ=]?B\;,"aGP¥a¥ 3�Þú ùñ��C� �9ÿSøM÷¨ù^ôjöZ�YôÙNc\ÚIìZ"=@N»_wFB×�àSF�FÂå^IMFÚ×tI@_'A"Cè?BC�Û¤CE?Â?BCETÄFÂ=AOØ'Iì_^IMVÙC@PFÂåSCE?ÂCtÛ¤=@?BC´NbFÆNc\¶_S=MFÆ_^CE×�Cs\Â\BIM?BØÙFÂ=ÙA"C%TYC�áD_SCET�?BCtZ"CEIKFBCETYa»Ø@Þ¦ Q�?OSa ¦ �¤¨§ �?a + [�©a�aª¡ Z£?¤a¤ aNfQ�?I�� �£¬«-�¤a¤

ïSÞ$i$��YE÷µö��>×�=@_D\lFB?Âà^×�FB\�\GZ"CE×tNbÛ¤Ø%FÂåSC�I@×�FÂN»=@_:FB=´ADC�FBI@L@CE_�IKFÅIy×�Ct?ÂFBI@N[_�ZD=]N[_wF ×�àYF),¤A"=@àS_DTFÂ=ÖFBåSCI]TYäON»×tC�3�Þ�߶åSCC®Q"¡?N¯?SP&"¡"f�¯ IM_^T°N¯f�a?èL@CtØOÜ�=@? TS\:TYCtá^_SC=1^Q��ù'FÂå^CAD=YTYØÚZ^IM?ÂF�=MÛ�FBåSC:I@TYäONc×�CyN»\¶CtâYCE×�àSFÂCETÜ�N[FÂå9?ÂCs\GZ"CE×�FFB=çFÂåSC´Z"=@N»_]F ×�àYFsÞO%>FÂå^Ct?BÜ�N»\ÂCFÂåSC:A"=YTYØÚ=@Û�I@_9I]TYäON»×tC´N»\¶ä]Ct?BØÚ\GN»VÙN[acIM?�FB=ÙFÂåSC:A"=YTYØ�=MÛJIJ>=KÜCEä@Ct?sPwÜC:\GFÂN»a[aqåDIä@Cy_S=MF¶Û¤=@à^_^T9IMZSZS?B=@Z^?ÂN[ãIKFÂC�VÙC�FÂ?BNc×%FB=w=]a»\>FÂ=ZS?BCE\ÂCt_wFyþwà^I@_]FBNbF IKFBN[ä]C�?BCE\ÂàSabF \>=]_ÖFÂå^Cç\GFÂ?Bà^×7FBàS? IMa;×�=@VÙZSa»C�âYN[FlØ9=MÛú$%ÇH'Z^?Â=]Ý@? IMV~\tÞ

%Ç_SC9ZD=w\Â\ÂN[A^a[C9?BCEI@\Â=@_ÄVÙN[Ý]å]F~A"C9FÂåSCÖacI@× L =MÛÇV�àSa[FÂN»Z^IM? I@TYN»Ý@V VçCtFÂ?BN»×E\çFÂå^IMFÚIM?BCäI@a[NcTÖ=]_èAD=@FÂå)=@A?�lCs×7FGã=@?BN»Ct_wFÂCsTÖIM_^TÖÝ]Ct_SCE?BIMFÂN»ä@C%ZDIM? I@TYN»Ý@V9ÞD߶å^Ct?BC�IM?BC�ZS?B=@Z"=]\BIMac\¶FB=VçCsI@\ÂàS?BC�\GZ"CE×tNbáD×�Û¤CEIMFÂàS?BCE\�=MÛÆú$%ÇH ZS?B=@Ý]?BI@VÙ\Ùí ²SP ³KðPqASàYFÙ=@à^?�IMZSZ^?Â=wI@× åìN»\%FÂå^IMF�N»_ZS?BI]×7FBN»×tC)IÄVç=]?ÂCì\ÂàSNbF IMASa»C VçCtFÂ?BN»×)å^I]\�FB=ÄA"CIMASa»CéFB=ÄVÙCsI@\ÂàS?ÂC \G=]àS_^TYa»Ø N[_ VÙ=@?BCZ^IM? I@TYN»Ý@V~\´IKF:=]_^×�C]Þq߶åSC~×�=@VÙZSa»C�âYN[FlØè=MÛ�IM_ìú$%ÇH ZS?Â=]Ý@? IMV TYCEZDCE_^TS\´=]_)FBåSC%;%ÇH×�=@VÙZ"=@_SCE_wFB\çIM_^T FÂåSCÖú$%ÇHJãj\GZ"CE×tNbá"××t=@_^\GFÂ?Bà^×7F \tÞ;߶åSCE?ÂCtÛ¤=@?BC~FÂåSCÖ×t=@VÙZSa»C�âYNbFlØ×t=@àSacTADC�\B×tIMFGFÂCE?ÂCsTèA"C�FlÜ�CtCE_éFBåSC�ú$%ÇHJãj\GZ"CE×tNbáD×çZ^I@?GF \J,¶N[_ìZ"=@N»_]F ×�àYFÂãTSC�á^_SN[FÂN»=@_D\tPqI]TYäON»×tCE\EPC�FB×@Þ 3�PSFBåSC%=@Af�lCE×7FÂãQ=]?ÂN»Ct_wFBCET�×�=@_D\lFB?Âà^×�FB\2,¿×�acI@\B\GCs\tP^N[_Så^Ct?BNbF IM_^×tC@PSCtFB×MÞM37P"IM_^T9CEä@CE_�N»_9FÂå^CZS?Â=Y×tCETYàS? IMa[ãj\lFlØOa»CyN»VÙZSa»CtVÙCt_wFBIMFÂN»=@_9=MÛ;FBåSC%VÙC�FBåS=YTS\tÞO>ÆCE_^×�C%N»_Ö=@à^?Æ=@ZSN»_SN»=@_ÖÜ�C:_SCtCsTFÂ=~IMZSZ^a[Ø�I�VÙCtFÂ?BN»×ÇFÂåDIKFÆVÙCEI]\GàS?BCE\�Ü�Cta»aqVÙ=@?BCÇZDIM? I@TYN»Ý@V~\¶IKF�FBåSC%\ÂI@VÙCÇFBN[VÙC]Þ

ëJâYZDCE?ÂN»Ct_^×tCE\%\ÂåS=KÜûFÂå^IMFçú$%ÇH ZS?B=KäON»TSCE\:IÖA"C�FÂFÂCt?ç\G=]a[àSFÂN»=@_ìÛ¤=@?�Iè×�CE?GF IMN»_ì\ÂC�F�=MÛZS?Â=]ASa»CtV~\,¤C@Þ Ý^ÞJa»=@Ý]Ý@N»_SÝ^P;TYCEASàSÝ@Ý]N[_^Ý^P;C�F ×MÞM37ÞÅæj_ÄFÂåSNc\ÙZ^I@ZDCE?~ÜC9N»_wä]CE\GFÂN»Ý]IMFÂCÚÜ�åDIKF~N»\×�=@VÙVÙ=@_ÚN[_ÚFÂåSCs\GCÇZS?Â=]ASa»CtV5Ý@?B=@àSZD\JFÂåDIKF¶?ÂCE_^TYCt? \JFBåSC´ú$%ÇHÄ\Â=@a»àYFÂN»=@_�N»_wFÂàSN[FÂN»ä@CEa[ØÙCsI@\GãN[CE?EÞ?8�IM_�Ü�C�á^_^TÚZS?B=@ASa»CtV5\GCtFB\+N[_�Ü�å^N»× åÚú$%ÇH ZS?Â=KäONcTYCE\+I%A"C�FGFBCt?¶\G=]a[àSFÂN»=@_[´$l åOØÙTS=ÜC¶\GCEC=]_SC�\G=]a[àSFÂN»=@_�CsI@\ÂN[CE?�à^_^TYCt? \GFBIM_DTSIMASa»CFÂå^I@_%FÂå^C�=@FÂåSCE?;NbÛDFÂåSCEØ�IM?BCN»VÙZSa»CtVÙCt_wFÂCsTà^\GN»_SÝyTYN��Ct?BCt_wF;Z^IM? I@TSN[Ý]VÙ\�´A>Æ=KÜ ×tI@_%Ü�C+ZS?B=Kä@CJFBå^IKF�Û¤=@?�FÂå^=]\ÂCIKÛ¤=]?ÂCEVÙCt_wFÂN»=@_SCsT:ú$%ÇHZS?Â=]ASa»CtV~\´FÂå^C\Â=@a»àYFÂN»=@_^\�IM?BC~_S=MF�=]_Sa»Ø)N»_wFÂàSN[FÂN»ä@CEa[Ø ASàYF�I@a»\Â=Ö=@Af�lCE×7FBN[ä]Cta»ØéA"C�FÂFÂCt?y´æj_

5

=@? TYCt?+FÂ=�I@_^\ÂÜCE?ÅFÂå^CE\ÂCÇþwàSCE\GFÂN»=@_D\tP]ÜCyIMN»VûFB=�I@_^IMa»ØahtCÆFBåSC(0´I@_SÝMã=MÛ¨ã�#S=]àS?$,�0y=N#c3�{ÇC�ã\GN»Ý@_�ZDIKFGFBCt?B_^\ÇíMWð"IM_DTÙFÂåSCEN[?¶N»VÙZSa[CEVÙCt_wFBIMFÂN»=@_^\N[_ÚZSàS?BC/\ÂZDCs×7F ÑÇÔGÍgtûÎf¸gu¹ovu-ÎYÏ^ÔGÓ

ëJâOFÂCt_DTYN[_^Ý)FBåSC�VçCtFÂ?BN»×?BCEþwàSN»?ÂCs\%FBåSCNcTYCE_]FBNbá"×tIKFBN[=]_ =@ÛÇú$%ÇHJãj\GZ"CE×tNbáD×ÚZS?B=@Ý@? IMV Cta»C�ãVçCE_wFB\EPwI@_^TÙFÂå^CtN»?V~IMZ^ZSN[_^Ý%FÂ=çIM_Úú1ÆãÝ@? IMZ^åqÞ@æj_Ú\ÂCE×�FÂN»=@_Úï�ÜC>å^Iä@CÆCE_OàSVÙCt? IKFÂCsTÙFÂå^CVç=w\lFÅN»VÙZD=]?GF IM_wF+ú>\ÂZ"CE×7F�×t=@VÙZSa»C�âYN[FlØ�=MÛ_^I@VçCsTÙZD=]N[_wF ×�àYF \ÅNc\;FÂåSCÇ\GàSV =MÛFBåSC>×�=]VçãZSa»C�âYNbFlØ =MÛyFÂåSCEN[?~_^I@VÙCE\,`�ÖAOØ TSC�ÛµIMà^abFy3ÙIM_^TÄFÂå^C�Z"=@N»_wFB×tàYF�NbF \GCEabÛlÞ߶åwàD\~NbÛyFÂå^CZS?B=@Ý@? IMVÙVÙCt?ÆTSC�á^_SCs\>I~×tCt?ÂFBIMN»_ÖZ"=@N»_]F ×�àYFsPD_DIMVÙCE\�N[FEPIM_DT9N[_^\GFÂCsI@T9=@Û;?BCtZ"CEIMFÂCsTYa[ØTYC�áD_SN[_^ÝÖNbF�I@Ý]I@N[_ ?ÂCtÛ¤Ct? \yFÂ=9FBåSC~ZD=]N[_wF ×�àYF�AOØ)NbF \:_^IMVÙC]P-FÂåSC�×t=@VÙZSa»C�âYN[FlØé=MÛ�FÂå^C×�=YTYC%×EIM_9ADC�?ÂCsTYà^×�CsT-ÞSæj_è\ÂCE×�FÂN»=@_èîçÜC:åDIä@C:\GCECt_qPSFÂå^IMF�FÂåSC%àD\ÂI@Ý@C´=MÛ;Û¤àS_^×�FÂN»=@_^\TYCE×t?ÂCsI@\ÂCE\�FBåSCÆ×�=]VçZ^a[CtâON[FlØ@P@ADCs×tIMàD\GC�AOØ%V~IMLON»_SÝ%IyÛ¤àS_^×�FÂN»=@_~×tI@a[aQPKFBåSC>I@T^TYCETÙ×�=]VçãZSa»C�âYNbFlØÙNc\+=]_Sa»Ø�FBåSCÇÛ¤àS_^×7FBN[=]_e \+_^I@VÙC@POIM_^T~N[FB\Z^I@?BI@VçCtFÂCE?B\EÞ]߶åSCÇà^\BIMÝ@C>=MÛq_^I@VçCsTZD=]N[_wF ×�àYF \�N»\�I@_^IMa»=@Ý]=@à^\FB=�FBå^IKFÆZ^?Â=Y×�CsTYàS?BC@Þ

ïSÞ$i$��YE÷µö���ó7P�Û¤?B=@V =]àS?�VÙC�FB?ÂNc×� \�ZD=]N[_wF�=@Û>äON[CEÜ/IM?BCÚASàSN»a[FçàSZ Û¤?Â=]V FlÜ�=èZ^I@?GF \tý�FÂå^CÛ¤àS_^×�FÂN»=@_9Z^IM?ÂFEPSI@_^TFÂåSC:Z"=@N»_wFB×tàYFÆZ^IM?ÂFEÞÈ ß¶åSCÇVçCtFÂåS=YTÙÛ¤=]?+VÙCEI]\Gà^?ÂN»_SÝ:FBåSC>Z"=@N»_wFB×�àSFZ^I@?GF+åDI@\+I@a[?BCEI]TYØ�A"CtCE_ÚTYCs\Â×t?ÂN»ADCsTN»_FBåSC:N[FÂCtVÉ�@Þ

6

È ß¶åSCZSàS?BCta»ØéÛ¤àS_D×7FÂN»=@_ Z^IM?ÂF�Nc\�I@\%Û¤=@a»a[=KÜÆ\EÞ�ú>_ I@TSäwNc×�CN \�åSCEI]TYCt?�N»\�a[N»L@C�FÂå^IMF=MÛ+IÚ\GZ"CE×tN»I@a�Û¤àS_D×7FÂN»=@_ \EPDÜ�N[FÂåèFÂåSC�L@CtØOÜ�=@? TS\Ê®Q"¡?�¯fOP9"¡"?N¯DPD=]?/N¯?�afÚI]\FÂå^Cè_^IMVÙC]PJÛ¤=@a»a[=KÜ�CET'AOØ FBåSCÖ?BCtÝ@à^a»I@?~Z^IM? IMVÙC�FBCt?~a»N»\GFEÞ�߶åSCÌË7_DIMVÙC"ËVÙN[Ý]åwFA"CZS?BCE×tCtCsTYCETìAOØIè?BC�FBàS?Â_ FlØwZ"C@Þ;߶åSCAD=YTYØ=@Û>IM_ I]TYäON»×tCÚTY=OCE\�_S=MFÙ\GCECtVTYN��"CE?ÂCE_wFyÛ¤=]?´FÂåSCÙZS?B=@Ý]?BI@VÙVçCE?>FBå^IM_)FÂå^C~AD=YTYØè=@Û�IÛ¤àS_^×�FÂN»=@_ Ü=]àSacT-Þ-ëÅä@CE_N»_IM_N¯?�af�I]TYäON»×tC@P]FÂåSCyL@CEØOÜ=]?BT ¦ ¯fQ�a"�TS=wCs\�_S=MF�\GCECtV/TYN��"CE?ÂCE_wFÛ¤?B=@VIM_ =]?BTYN»_^I@?ÂØìÛ¤àS_D×7FÂN»=@_ ×EIMa»a¿ÞÅ߶åSCE?ÂCtÛ¤=@?BC�FÂåSC9Û¤àS_D×7FÂN»=@_ Z^I@?GF� \~×t=@VÙZSa»C�âYN[FlØ N»\VÙCEI]\GàS?BCETÚFÂå^C%\ÂI@VçC´Ü¶IØ�I@\)NbÛJFBåSC�A"=YTYØ9Z^IM?ÂFÇ=@Û;FÂåSCÙI@TYäONc×�C%Ü�=@à^a»T9A"C�N»_ÖFÂå^C�ótö øBÿQ�ç=MÛJFÂå^C�ZS?BCETYNc×tIMFÂCs\TYC�áD_SCET AOØÄFÂåSCèZ"=@N»_wFB×tàYFEÞ)8=@VÙZSa»C�âÄZ"=@N»_wFB×tàYFTYCtá^_SN[FÂN»=@_^\ÚADCEå^Iä@CÖa»N[L]CÖ_SCE\GFÂCsTZS?BCETYNc×tIMFÂCE\EÞq߶åOà^\´FBåSC�×�=]VçZ^a[CtâON[FlØè=@Û¶I@_ I]TYäON»×tCçNc\´FBåSC�×�=]VçZ^a[CtâON[FlØè=@Û¶I]TYäON»×tC� \AD=YTYØV�àSa[FÂN»ZSa»N[CsT�AwØÚFÂåSC:×t=@VÙZSa»C�âYNbFlØÚ=MÛ;NbF \�ZD=]N[_wF ×�àYFsÞ

ü^Þ$iyóQÿQ�Bötô¿ó)IM_^T ×�acI@\B\GCs\�å^Iä]C)I a»=MF�N[_ ×�=]VçVÙ=]_ Û¤?Â=]V�FÂåSCì×�=]VçZ^a[CtâON[FlØ'ZD=]N[_wF9=MÛäON[CEÜ:Þê�=@FÂåéV~IØ9N»_^×�a»à^TYC~TSIMFBISPIM_DTèVÙCtV�ADCE?ÇÛ¤à^_^×7FBN[=]_^\tÞq߶åOà^\>FÂå^CE\ÂC�VÙCtV�ADCE?B\=MÛSI@\ÂZ"CE×7F \q×tI@_´ADCVçCsI@\ÂàS?BCET>FÂåSC+\BIMVÙCÅܶIØ>I]\-NbÛOFBåSCtØyÜCE?ÂCJN»_%×�acI@\B\ÂCE\EP Û¤=@?qFBåSNc\-FÂå^CVÙC�FÂå^=OTéN»\´TYCs\Â×t?ÂN»A"CETèN»_ îSÞ"úÇ\GZ"CE×�FB\´V~IØÖIMac\G=Úå^Iä@C�VçCEV�A"Ct? \>=@Ûú$%ÇHJãj\GZ"CE×tNbá"××�=]_^\lFB?ÂàD×7FB\EÞ?l C:å^Iä@C´×�acI@\B\GN[á^CsT�FBåSCE\ÂC%×�=]_^\lFB?ÂàD×7FB\�N»_wFÂ=ÙFlÜ�=�Ý]?Â=]àSZ^\EÞÈ ß¶åSCç×�=@VÙZSa»C�âYN[FlØ=@ÛÇñ���YE÷µö���ó PqI@_^Tìù"ñ��J����ÿ^øK÷¨ùDôö��Yôµó%Nc\>FBIML]Ct_ÖN»_wFÂ=Ú×t=@_^\ÂN»TSCt?ÂãIKFBN[=]_ Ü�åSCE_ VÙCEI]\GàS?BN»_SÝÖFÂå^C�I]\GZ"CE×�FEÞJ߶åSCE\ÂC×t=@_^\GFÂ?Bà^×�FB\çTYN[?BCE×�FÂa»ØI��CE×7FçFÂå^C

ܶIØ%FÂå^CÆZS?B=@Ý@? IMVÙVÙCt?J\ÂCtCs\JFÂå^C>×�=YTYC]Þ]WOåSCÆ_^CtCET^\JFÂ=�àS_^TYCE?B\GFBI@_^T�FÂåSCs\GCÆVÙCEV�ãA"Ct? \>FB=ÚA"CçI@ASa»C%FÂ=9×�=]VÙZS?ÂCEåSCt_DT9FÂåSCÙ×�=]VÙZSa[Ctâè×�=@_D\lFB?Âà^×�FyTYCs\Â×t?ÂN»A"CETÖAOØ9FÂå^CZS?B=@Ý]?BI@V�Þ

È ú>\�=@Û%_S=KÜ N[_wFÂCE?GãQFlØOZDC TYCE×ta»I@?BIMFÂN»=@_^\Úa[N»L@C½��� öZ*»ñKõ+�ÿ^ñKõ ��ùDôµó7P;��� öZ*»ñKõ+�°�tõ7õÂøMõ ó P��� ö�*[ñKõ �x+ñMõ7ù^÷¨ùa�MP^I@_^T=MFÂå^Ct? \�IM?BCÇ_S=@F¶FBIML]Ct_�N[_wFB=ÙI]×t×t=@àS_wF�Ü�åSCt_�VÙCEI]\GàS?BN»_SÝ×�=]VÙZSa[CtâYNbFlØ]Þ?l)C%×t=@_^\ÂN»TSCt?�FÂåSCs\GC:I@àYâYN[a»N»I@?ÂØ�×t=@_^\GFÂ?Bà^×�FB\�N[_ÖúÇ\GZ"CE×�F IM?BC�\Gà^VçVÙCsT9àSZèÜ�N[FÂå9FÂå^C�×�=]VÙZSa[CtâYNbFBN[Cs\Æ=MÛ;FBåSC�TSIMFBISPIM_^TFBåSC:VÙCtV�A"Ct?¶Û¤à^_^×7FBN[=]_^\�=MÛ�FÂåSC%I]\GZ"CE×�FB\EÞ

l)CÆåDIä@CÆ\ÂCtCt_~N»_�\ÂCE×7FBN[=]_�îyFÂå^IMFÅFÂå^CÆäON»\ÂN[A^N[a»NbFlØç=MÛq×ta»I]\Â\ÂCE\EPMIM_^TÙN[FB\+VÙCtV�A"Ct? \+TS=wCs\J_S=@FN[_?wDàSCt_^×tCÙFÂåSCEN[?%×t=@VÙZSa»C�âYN[FlØ@Þe#S=]?´FÂåSC�\BIMVÙCç?ÂCsI@\Â=@_^\yÜ�C~TY=9_S=MF:F IML@C�FÂå^N»\%IMFGFB?ÂN»ASàYFBCN[_wFÂ=~×t=@_^\ÂN»TSCt? IKFÂN»=@_N[_�FBåSC%×tI]\GC´=MÛ;I@\ÂZDCs×7F \tPSI@_^T�NbF \�VÙCtV�A"Ct? \¶CtN[FÂå^Ct?sÞ

Í ÎÏuOÒ$qlÎf

ß�=)äI@a[NcTSIMFÂCÚ=]àS?çVçCtFÂ?BN»×ÚÜC�å^Iä]CÚ× åS=]\ÂCt_ìFBåSC0y=N#¼{ÇCE\ÂN[Ý]_ HJIKFGFBCt?B_^\��,GíMWðb3�N[VÙZSa»C�ãVçCE_wFBIKFBN[=]_^\J,líV�tü@ðb3>Û¤=@?´VÙCEI]\Gà^?ÂN»_SÝ^Þ%Ç_SCç=MÛFÂåSCÙ?BCEI]\G=]_^\>ܶI@\ÇFBå^IKF:N»_ÄíV�tü@ð;Ü�C�á^_^T IÛ¤àS_^×7FBN[=]_^IMa»a»Ø CsþwàSN[äKI@a[CE_]F�N[VÙZSa»CtVÙCt_wF IKFÂN»=@_ =MÛÆCEI]× åZ^IKFÂFÂCE?Â_ N»_ ú$%ÇH�I@_^TÐ%;%ÇH;Þ;ú�FFÂåSC´\ÂI@VÙCÆFÂN»VÙC@P]FÂåSCy?BCt_S=KÜ�_SCsT~IMàYFBåS=@? \ADCEåSN[_DT~FÂå^CÇN»VÙZSa[CEVÙCt_wFBIMFÂN»=@_^\a[CtFà^\¶I]\Â\ÂàSVÙCFÂå^IMF�FBåSCI]\GZ"CE×�FGã=@?BN[CE_]FBCET FBCE× åS_SNcþwàSCE\�Ü�Ct?BCÚå^I@_^TYa»CET ×t=@?B?ÂCs×7FÂa»Ø@Þ1l)C9IMac\Â=é\Gà^ZSZD=w\GCsTFÂå^IMF�{ÇH+\¶IM?BC�_SCtàYFB?BI@a^FB=�×�?B=]\B\B×�àYFÂFÂN»_SÝ�×�=]_^×�CE?Â_D\tÞ�l CÇTSN»TÚ_S=MF× å^=w=w\GC>C�âSIMVÙZSa»CE\ÅFÂå^IMFIM?BCÇÜ�Cta»abãLO_S=KÜ�_×t?Â=w\Â\B×�àSFGFÂN»_SÝçZS?B=@ASa»CtV~\F,¤C]Þ ÝDÞYa[=]Ý@Ý@N»_SÝDP@FB?BI]×�N»_SÝ^POC�F ×�3�PSASàYFÆVÙ=@?BCyÝ@Ct_SãCt? IMa-=@_^CE\EPOFÂå^IMFF�ç÷��S^Oô�A"CyN»_�FÂåSNc\�ZS?B=@A^a[CEV \GCtFEÞ

7

ÕèIM_OØÚZ"Ct=]ZSa»C´FÂåSN»_SLÖú$%ÇH ?ÂCsTYà^×�Cs\¶FÂå^C�×�=]VÙZSa[CtâYNbFlØ=MÛ�FÂåSC�TSCE\ÂN[Ý]_ÖZ^IKFÂFÂCt?B_^\�SN»V�ãZSa[CEVÙCt_wFBIMFÂN»=@_^\>ADCs×tIMàD\GC�=MÛJFÂå^C�Z^IMFGFÂCE?Â_D\�×�?B=]\B\Â×tàYFGFBN[_^Ý�IMZSZS?B=]I]× åSCE\EÞO%ÇAOäON[=]à^\Âa[Ø]PSFÂå^CTYCE\ÂN[Ý]_ìZ^IKFÂFÂCt?B_^\�å^Iä@C~A"CtCE_ ×�?BCEIKFBCETìI@\�\G=]a[àYFBN[=]_^\´Û¤=]?%FÂå^C�_S=@_Sã¿FB?ÂN»äON»I@aÅZS?B=@ASa»CtV~\:N»_%;%ÇH�Þ߶åSCÙIMZSZ^?Â=wI@× å9=MÛú$%ÇH ×tIM_)TYCE\B×�?BN[A"C%FBåSCE\ÂCÙ\G=]a[àYFBN[=]_^\ÆCsI@\ÂN[CE?ÇAOØÖú$%ÇHg \>_^CtÜa»I@_SÝ@à^I@Ý@C´×�=]_^\lFB?ÂàD×7FB\EÞ

߶åSCé\GFÂ?Bà^×7FBàS?ÂCÖ=@Û´FÂåSCs\GCèN»VÙZSa»CtVÙCt_wFBIMFÂN»=@_^\ÚN»\I@\ÙÛ¤=]a[a»=KÜÆ\EÞÅë+I@× å ZDIKFGFBCt?B_ å^I@\ÚI\ÂZDCs×7F FBåSC9Z^IKFÂFÂCt?B_^\EÞJú>\ç=MÛy_S=KÜ6=@à^?VçCsI@\ÂàS?BCtVÙCt_wF�FÂ=O=]aqN»\>IMASa»C´FÂ=�ZDIM? \GC:IM_DTVÙCsI@\ÂàS?ÂCC�EüÙ=@àYF>=@ÛJî@ï~TYCE\ÂN»Ý@_9Z^IKFÂFÂCE?Â_9N»V�ãZSa[CEVÙCt_wFBIMFÂN»=@_^\EÞ^߶åSC:F IMASa»C�\ÂåS=KÜÆ\¶FBåSC�ú1ÆãVçCtFÂ?BN»×´äKI@a[àSCs\tP"IM_^TFBåSC�C��"Cs×7FÂN»ä@C:a»N»_SCE\>=MÛ×�=YTYCC,¤ë�.H%;8Ê3�ZDCE?ÆZ^IMFGFÂCE?Â_D\tÞ

{ÇCE\ÂN[Ý]_9H;IMFGFBCt?B_ æjVçZ^a[CEVçCE_wFBIKFBN[=]_ ú8=]VçZ^a[CtâON[FlØ ë-�"Cs×7FÂN»ä@C2.H%;8"? ¦ ?N¯ �GIäKI WNW îNW

ú$%ÇH �f� î@î®?¯O� « �GIäKI î@ïN� W��

ú$%ÇH î@ï"W W�³®a9�NaN¯ �GIäKI îf��³ ���

ú$%ÇH î��f� ²�²?f�N¯fN?N¯ �GIäKI ³Q� ïMü

ú$%ÇH ³]ï î��¡?Q�f�¯aÑaÒfNN¿Q" �GIäKI �N�Eï �Kü

ú$%ÇH �î�³ ²"W¡f�Ñ"ÀQ[ « ¿? �GIäKI î�³�³ ²�²

ú$%ÇH î�»�² Wa��??N¯ ¦ ¯f�fN¯ �GIäKI ��²"W �����

ú$%ÇH ��²"W ���EïÓ Ó �af �GIäKI ³N³ ï@ï

ú$%ÇH ��»�² üaW�®IÁ��¯ N¯ �GIäKI ïaWü ³@ï

ú$%ÇH ïN�N� »"W¦ ¯?NfNaÑ ¦ �GIäKI ��»"W �Mï

ú$%ÇH î��Mü ��²Á�fN? �GIäKI îN��³ ³"W

ú$%ÇH ��W�³ ���@ïÁ�a¯f�f « Ñ �GIäKI î�²N� ��²

ú$%ÇH WMï]î ²�»f Óf¦ "Nf�ÒfN"¿QN �GIäKI �S��» ü]ï

ú$%ÇH �S��» ü"� "Á?S?N¯ �GIäKI ïN��� »@ï

ú$%ÇH ïN²]î »N�

8

ú/×�=@VÙZ^I@?ÂNc\Â=@_èA"C�FlÜ�CtCE_éFBåSC~N[VÙZSa»CtVÙCE_]F IKFBN[=]_)=@ÛFBåSC~TYCE\ÂN[Ý]_)ZDIKFGFBCt?B_^\´N»_)FBåSC%;%ÇHIM_^T ú$%ÇH5Ü�IØÄ×tIM_ A"CÖÛ¤=@à^_^T N[_2íV�tüDP���²MðQÞÅ߶å^CE\ÂC�ZDIMZ"Ct? \~C�âYZSacIMN»_'FÂå^IMF=�SW)=@àSF�=MÛîMï�Z^IKFÂFÂCE?Â_^\%å^I@T)C�âYåSN»ASN[FÂCET\Â=@VÙC~TYCEÝ@?BCtCÙ=MÛ�×�?B=]\B\Â×tàYFGFBN[_^Ý^Þ-߶åSCtØ IMac\G=ÖTYCs×�acIM?BCçFÂå^IMFN[VÙZSa»CtVÙCt_wFBN[_SÝéFÂåSC�Z^IMFGFÂCE?Â_D\�N»_ ú$%ÇH�åDI@\%V~I@_wØìA"Ct_SCtáSFB\EPJIMVÙ=@_^ÝÖFÂåSCEV FÂåSCVÙ=]\GFN[VÙZ"=@?ÂFBIM_wF>A"CtN»_SÝÚFÂåSC�I@ASN»a[N[FlØ�FÂ=�a»=Y×tIMa»NVhtC:FBåSC�×�=YTYC�Û¤=@?yI~Ý@N»ä@CE_�ZDIKFGFBCt?B_qÞÕèIM_OØÚZDIKFGãFÂCt?B_^\�×EIM_'ADCéN[VÙZSa»CtVÙCt_wFBCET I@\�Iì\ÂN[_^Ý@a»CèI@\ÂZDCs×7FEP+=@?�I@\�îì×ta[=w\GCEa[Ø ?ÂCEa»IMFÂCET'I@\ÂZ"CE×7F \tÞ%ÇàS?ÇVÙCtFÂ?BN»×�CE\ÂZDCs×�NcIMa»a[Ø9?BCtܶIM? TS\>×t=OTSC�a»=O×EIMa»N�hsIKFBN[=]_qÞD߶åSCk%;%ÇH ä]Ct? \GN»=@_^\Ç×tI@_è_S=MF´A"CI@\:Ü�Cta»abãj\GFÂ?Bà^×7FBàS?ÂCsT)I]\yFÂåSCú$%ÇH ä]Ct? \GN»=@_D\tPÜ�åSCE?ÂCÙFBåSC�×�=YTYC~Nc\:VÙ=@?BC~V~IMN»_]F IMN»_^IMA^a[C]PIM_^T~×t=@VÙZS?BCtåSCE_^\GN»ASa»C@Þ@ú>_S=MFBåSCt?N[VÙZ"=@?ÂFBI@_]FADCE_SC�áSFN»\ÅFÂåSCy×�=YTYC� \J=]ASa[N»äON[=]à^\Â_SCE\B\tÞ]߶åSN»\ADCE_SC�áSF~?BCE\ÂàSa[FB\ÙTYN»?ÂCs×7FÂa»ØìÛ¤?B=@V a»=O×EIMa»N�hsIKFBN[=]_qý�I]\�FÂåSC�Z^IKFÂFÂCE?Â_'N»\ça»=Y×tIMa»NVhtCET N[_ IM_'I@\GãZDCs×7FEPwN[F�TY=OCE\+_S=MFN[_OäKI@TYC>NbF \ÅZ^I@?GFBN»×tN[Z^I@_wFB\EÞ�>ÆCE_S_SCtV~I@_ÚIM_^TCÔ´Nc×ZhsIMa»CE\Å\GFBIMFÂCsTçFÂå^IMF+FÂå^Cú$%ÇHä@Ct? \ÂN[=]_^\;IM?BC�Vç=]?ÂC�VÙ=YTYàSacIM?+AwØCWü?Õ�P]IM_^TçVÙ=@?BC�?ÂCEà^\ÂCEIMA^a[C¶AOØJ�]îNÕÚÞMú�×t×t=@? TYN»_SÝFÂ= í��EüMðJ\G=]VçCçZ^IMFGFÂCE?Â_D\�N»VçZ^a[CEVçCE_wFBIKFBN[=]_éV~IØÖTYNc\ÂI@ZSZDCsIM?yN[_wFB=ÚFBåSCç×t=YTYC�A"CE×EIMà^\ÂCç=MÛú$%ÇHg \�×t=@_^\GFÂ?Bà^×�FB\F,¤C]Þ ÝDÞOFÂåSC:TYCs×�=@? IKFB=@?�Z^IKFÂFÂCE?Â_[3�ÞS߶åSNc\�×tIM_�IMac\G=ça»CEI@TÚFÂ=~×t=@VÙZSa»C�âYNbFlØTYCE×t?ÂCsI@\ÂC@Þ

æj_ \G=]VÙC×tI]\GCs\%Ü�C×EIM_Ä\ÂCtCÚFÂå^IMF�FBåSC�ú$%ÇHûN[VÙZSa»CtVÙCE_]F IKFBN[=]_ Ü�I]\�a»CE\B\ç×�=@VÙZSa»C�âAwØ=@àS?�VÙCtFÂ?BN»×´CEä@Ct_�N[Û�FBåSC:ë�.H%;8 _OàSV�ADCE?ÆÜ�I]\¶Ý@?BCEIMFÂCt?sÞY߶åSCE\ÂC%IM?BCyFÂåSC%×EI@\ÂCE\¶Ü�åSCE_à^\GN»_SÝ´ú$%ÇH)ܶI@\;I]TYCEþwà^IMFÂC@PsFBåSCE\ÂC�IM?BC+FBåSCgN? ¦ f�¯^PS®"9�N?�¯^P��®IÁ�N¯ N¯YPMI@_^TkÁ�fNfZ^IKFÂFÂCt?B_^\EÞ�>Æ=KÜ�Ctä@CE?EPsÜ�C�×EIM_Ù\GCEC�IÇ_OàSV�A"Ct?J=@Û^Z^IKFÂFÂCE?Â_^\�Ü�åSCE?ÂC�FBåSC�ú$%ÇHéN»VçZ^a[CEVçCE_wFBIKãFÂN»=@_^\JÜ�Ct?BCÆIKFÅa»CEI]\lF+I]\Å×�=]VÙZSa[CtâçI]\;FÂåSCEN[?�

â@ÃÆuvrEzwe uhle7|@n]>JÃbÇʶg�hl ubg@ ³ Ãv gEmKiyç ³ m]mOgM>}Ãvgä�zwve {j7gMÆÇÃv>xDz] e mOgM>}ÃÝÜ ³ {j{je {j{uvm]r ³ {jfYe Âij ¸ hluve mKile7| ³ hjiluo¨gtBil{l|QÜaÞ"n7ºJgthG|@{�g;ÞDnKns zwf]fYnEhjile7|kè�zwgEmKiluiGgtiluvegßçeÂil¼wnM|^è ³+¸�¸ ÖÉ�nEh`æM{j¼wnEfD]�} ¸�¸ p-OÊ�bgE{jrEn�º�OfwfDÃ"��átâ�àMQ�����S�

MÃ>ʶgtk:k%g@^JÃv"e vkÙfã>ÃvDÈsns¼]mw{jnEmD[ã>ÃvQ�vuv{j{jub|]e { "È]ÃÝÜ"d�e {juvrsm9p-g�ijileBhlmw{�é��ve k:e mKil{¶ntoã+e z]{lgEÀwve ¸ ÀMÁeBBi ¸ hlubeBmMile | nto»i¿ºJgthleE ³ |w|@ub{jnEm]¤É9eB{jbeÂK^�à�à��

á@Ã>ÊhGgE|]e ]æMu¨OÈ]à dyÃvOxDe {juve ]æMu¤NêÇÃÝÜaßÙgE{ileBhluvm]r ³ {jfYe BiGÈ@]ÉuvveBKa�������à@ÃÆÈEe7gEm@6ë�e {�Ê�z@Knsk%g�hl�ç ¼D�ë�gEm]m]QʶgEeBOÊ�z] ¼w m]e zw�Ü ¸ m´il¼weJ¿k:fOgtBi�nEo ³ {jfYe Âij ¸ hluve mKile7|

p-hlnsrthGgEk:k:uvmwr~nEm ¸ À@ÁQe Bij ¸ hluve mKile7|=ßçeBijhluv { 1è ³+¸�¸ ÄÉ�nEh`æM{j¼wnEfD�} ¸�¸ p�Ê�bgE{rEn7º�Ofwf^Ã�����6�Mf�������

��@Ã��gthjhluv{jnsm^EÉ'à ³ ÃvSßÙgtrse ¨�ä´Ã GÃv ³ }�nsk:f]be¾]ui¿/ßçe7gE{jz@hleÅ;gE{je7|>nsm/ê�e {iluvmwr�xDe se ¨ ³ }cß uvrsfwbgtmJê�ntilubBe { vyâ]·_�s¹ÂOf]fDà â

EsÃ�n7ºJgtiji7EÈ]Ã É Ãvs;g�æseBh7 ³ à x�ÃÝÜ�ãuvrsnthlnsz]{�d�e ìOm]uviluvnEmygtmO| ³ mOgtv@{juv{-nEoYpqhlnErEhGgEk }�nEk:fwveB¾Mui¿Xßçe7gE{jz@hle {�Ü ³ m�-¾]gEk:fwve´{juvmwrKêe {iluvmwr@IÞ;¼]e:Èsnsz@hlmOgt�nto @{ile k:{�gtmO| nEo[iQº;gthle��@]fwf^Ãb���à�Gy���Myà�á�à

y�MÃ)ä�uv gEve { MÊÇÃvsx"gtk:fwuvmwr@sÈ@ÃbSßçe mw|]¼we æsg�h7 ³ Ãv�ßÙgte7|wgMM}Ãv�ub|]e uhGg¶x^nsfYe { }Ãvx^nsuvmwrtilubeÂh7ÈEe7gEm@�ßÙg�hls]¿hjºÅuvmDwÈ]ÃÝÜ ³ {jfYe Bij ¸ hluve mKile7|�p-hlnsrthGgEk:k:uvmwr@]�} ¸�¸ pwubm]bgEmO|S f]hluvm]rseBhj

�eBhlbgEm]r´xQê�} nE¨ÃDy���]EOf]fDÃ"�����t����S�M�à�à

��@Ã)ä�uv gEve { ]ÊÇÃvM�uvv{l|wgtbeEwJÃvK�z]rsz]mwuvmDMÈ@Ãb�ä�eÂhl{ile mD�ßÃv@p-gtbkÙKÈ]Ãv]Êhlub{º�nsb|^MÉ Ã ÊÇÃÝÜ ³ m¸ seBhlMubeº nEo ³ {jfYe BiGÈ]Ox[ê�} ns¨Ãa���s��Mwfwf^Ãa���s������M?�����@

�]Ã)ä�uv gEve {�ÊÇÃvw�e m]mwe k%gtmDwÈ]ÃÝÜYd¶e {juvrsm�p-g�ijileBhlmÙ¿k:fwve k:e mKiGgtiluvnEmçuvm�Èg7sg:gtmO| ³ {jfYe BiGÈ]¸�¸ p x ³ Mf]fDÃ^yâ@Âl���MQ�����S�

y�MÃ)ä�uv gEve { MÊÇÃÝÜ ³ {jfYe Âij ¸ hluve mKile7|ÇpqhlnErEhGgtk:k:ubm]r] ³+¸ p�}�nEk:fwz]iluvm]r{jz@hleB@{��á@·¨e {G¹Âwy���Ef^Syà�à�â

�â@ÃÆx^e {juve ]æ@u¨ êÇÃÝÜ qmw¼wgEmwBe |@e {juvrsm fwgtijileBhlm]{ ºÅuil¼ ³ {jfYe BiGÈ] cߣKE1í�î�î�ïSïSï�¢µªK �ðK K¡s�@ �ñ@§D¢¤@ s�îE �òK �ñt]K¡Sî�ó�ôSõ

7MÃÆpquº;n7ºJg�hl{�æMu¤ãÆà ÅÃÝÜ ³ ê�e {iluvm]r�x^e e Å}�nEk:fwveB¾@uiQ@ßçe gE{jz]hleE ³ }cß uvrsf][gtmPênEiluv e { 7@·_às¹Âwf]fDà ���t����@Dyà�á��

�á@ÃÆÈsgEk:e {y�Ãqpn7º;eBh7q�hlu[gtm ³ ÃeßÙgtbvn7fÜ ³ k:eBijhluv {Ç{jz]uile�o»nEh´rthGgEk:k%g�hj¿ÀwgE{je7|9{jnto»i¿ºJgthleEÃÈEnsz]hlmwgE^nEo nEo»i¿ºJg�hle$ßÙgEuvmKile mOgtmw e´�â]·_â¹+Ü"�S����6����â�·6�������K¹

�à@ÃÆpnth`ænsbtÀDföÃv uvbKeEa÷ÆÃ�Ü"Þ"n7ºJg�hG|]{gykyz]ilubfwgthGgE|]uvrsk nEk:fwveB¾@uiQ�k:e7gt{jz]hles"è ³+¸�¸ É�nEh`æM{j¼wnEfD]�} ¸�¸ p-OÊ�bgE{jrEn�º�OfwfDÃv����El ���Mf�������

���@à ¼]k:ub|]k:e uveBh7 ³ Ãv�¶gEm]e mMÀYeBhlr Ãv��mwbgtmO|^ãÆÃÝÜqQk:fwve k:eBmMiluvm]rä�m]n7ºÅmì}�nEmw e f@il{:uvm³ {jfYe BiGÈ]a�������

�KsÃ>ÊhlBrsnEhj eÂhlnsmKi7OßçubrEzwe �xDnEfYe s[�gtbÂhlube�p-gtzwvzw{ [ê¶g7Áu�¶gtÀ]hGg�Ü ¸ mil¼]eFã+e bgtiluvnsm]{j¼wuvfÀYeÂiQº;eBe mÙ}�@ vnEk%gtiluvÆ}�nsk:f]veB¾@uvi¿�gEmw|%il¼]e�d¶e rthle e¶nto ¸ À@ÁQe Bi ¸ hluve mKiGg�ilubnEmD"è ³+¸�¸ É�nEh`æM{j¼wnEfD]�} ¸�¸ p-OÊ�bgE{jrEn�º�OfwfDÃS���àtlE�

���MöÉ9gs|@beÂh7pÃÝÜ9Þ;¼]e:eB¾@f]hle {j{juvnsm�f]hlnsÀ]ve kÙpnE{ile7|�nEm�il¼we%Èg7Eg~Ê�e m]eBhluv uiQk%gEuvvubm]r~vub{i7�à�à�á

���@öÉ�eB@z"æeÂh7Jà È]ÃÝÜ��EgEvzOg�iluvmwrÚ{jnEo»i¿ºJg�hleç nEk:fwveB¾@uiQk:e7gE{jz@hle { -��ÞDhGgEm]{ à nEo[iQº;gthleqmwrEuvmwe eBhluvm]r]Osns¨Ãv��]wfwf^Ãb��S�E�ly��âS�M"yà�á�á

�Z�]ÃÆÈsg�Eg�sà �K £ME1í�îSîSøs�ð-¢j§7¥�¢�ùs�úaîtªK �ðK K¡s�@ �ñ�îtK �ù7£w�ùsK¡�ûSñK@�ùs¡s @§�î�ñ K¡ M§E @§�îSøSüK§s "ýyþ���MÃÆq vuvfw{jesà nthlr´ocnEhlk%g�iluvnsmD £MEí�îSî�ïSï�ï�¢Q �ùs¡7w§E -¢Q�ñ©�î�ñ©�îK7KªK Sòq¢¤£M�úw¡

10

On the Influence of Practitioners’ Expertise in Component Based Software Reviews

Miguel Goulão1, Fernando Brito e Abreu1

1 QUASAR Research Group, Centro de Informática e Tecnologias de Informação Faculdade de Ciências e Tecnologia, Universidade Nova de Lisboa, Portugal

{miguel.goulao, fba}@di.fct.unl.pt http://ctp.di.fct.unl.pt/QUASAR/

Abstract. Objective: To assess the influence of practitioners’ expertise in code inspection of software components. Method: Subjects expertise was determined based on their independently assessed academic record. Inspection outcome was represented by the diversity of defects found at two levels of abstraction. Re-sults: Statistically significant correlations among expertise and inspection out-comes were found in several cases . Conclusion: The effect of expertise is ob-servable in the inspection outcome and thus can be used in for software quality management purposes.

1 Introduction

Software development team members’ expertise is an essential factor to the success of a software project. Skilled developers are likely to produce better software than less skilled ones. Good code reviewers are likely to detect more defects than bad ones. In this paper, we assess the impact of practitioners’ skills in the context of code reviews performed on component-based (CB) software. The reported results are part of a wider experiment, briefly described in the following section, where we assess pract itioner’s performances in the development, quality control and integration of software comp o-nents, and compare them with an independent assessment of their skills .

A code review is a peer review of source code intended to detect defects before the testing phase begins , thus improving overall code quality. There are a number of code review processes being used in industry. Fagan inspections [1, 2] are considered semi-nal in this area. Other processes have emerged since then, that try to lower the costs involved in code inspections without sacrificing their benefits. Examples include con-ducting inspections offline, thus skipping the inspection meeting [3], or performing phased inspections, where the inspectors focus on a specific class of defects [4], although the latter technique has been criticized for being more costly than conven-tional inspections [5].

Understanding what drives inspections’ success has been a long time concern in the software community. Based on data collected from over 6000 inspections, Weller

11

studied the impact of the inspection process on software quality [6]. Among several other remarks, he pointed to the familiarity of the inspection team with the artifact being inspected as a key factor in inspection success. We may regard this as kind of domain expertise. Siy observed that while structural changes were largely ineffective in improving the results of inspections, the inputs for those inspections (the reviewers and code being inspected) were far more influential in the inspection outcome [7]. These findings were further explored in [8], to conclude that better inspection tech-niques, rather than processes, were the key to improving inspection effectiveness. Biffl and Halling combined reviewers’ expertise measures (software development skills, experience and an inspection capability pre-test) with different code inspection tech-niques [9]. While they could not find significant relationships between development skills and experience and inspectors performance, they found the inspection capability pre -test useful to optimize the inspection outcome by selecting ideal inspection teams. They also identified performance differences related to alternative code reading tech-niques , a result that is consistent with the findings of Laitenberger and DeBaud, in their systematic review on code inspections reading techniques [10]. Sauer et al. iden-tified individual’s task expertise as the primary driver of review performance [11].

In a totally different context (social psychology), Kruger and Dunning observed that the skill of a person in performing a task is closely related to the required skill to assess his own performance in the same task [12]. If we instantiate this insight into code production and code reviewing, we would expect the best programmers to also be the most effective code reviewers.

2 Problem statement

Our global goal is to analyze the outcome of a CB development process, for evalu ation purposes , with respect to the impact of practitioner’s expertise on defect introduction and detection, from the point of view of a project manager (in this case, the research team), in the context of an academic simulation of a component marketplace.

In this paper we are concerned with the impact of practitioners’ expertise in the out-come of CB software code reviews performed at the component level. We are seeking evidence on possible causal relationships between the expertise of practitioners in-volved in the code inspections and the diversity of defects reported during those inspections (Fig. 1). All inspections were carried out by a review team (RT) which included the development team (DT) and a peer team (PT). PTs consisted of develo p-ers of a different component, participating as independent code reviewers. The remain-ing process and product inputs are fairly similar for all code inspections, to minimize possible confounding effects.

We consider four potential causal relationships. The expert ise of the development team (1) may have a negative effect on the diversity of defects found. The rationale is that expert developers tend to introduce fewer defects on their code. Conversely, peer reviewers expertise may have a positive effect on the defects diversity (2). A similar rationale leads to the possible causal effect between the expertise of the review teams

12

as a whole (3) and defect diversity. Finally, we consider the difference of expertise between the developer team and the peer team (4) as a negative effect on defect diver-sity. If the expertise of the developers is higher than that of their peers, defect diver-sity is expected to be smaller than when the opposite occurs.

Review Team (RT)

Development Team (DT)

Peer Team (PT)

Defect Diversity

(3) RT Expertise [+]

(1) DT Expertise [-]

(2) PT Expertise [+]

(4) DT Expertise - PT Expertise [-]

Fig. 1. Exploring the impact of practitioner’s expertise in the outcome of a review process.

3 Experiment planning

3.1 Context selection

This experiment occurred in the context of a Software Engineering course held at the Universidade Nova de Lisboa, during the Spring semester of 2005. This course is of-fered on the 8th semester of their 5-years informatics degree. According to the new harmonized academic curricula adopted in Europe (Bologna model), these are 2nd cycle degree (MSc) graduate students . The course’s project consisted in developing a CB elevator system simulator from requirements definition to final product delivery. The programming language was Java, well-known to all subjects in the experiment.

Among other activities, the development process included a Fagan inspection per-formed on all the developed components . While PT members were knowledgeable in the inspected code basic requirements, they were not developing an alternative imple-mentation of that same component, to avoid biasing their review, besides better repro-ducing industry practice.

Standard Fagan inspection roles were assigned to the four RT members. The DT members got the moderator and author roles and the PT members the remaining ones (reader and recorder). An extensive checklist of common defects in Java programs was distributed (and its contents explained) to all RTs before code inspections took place. This paper focuses on the analysis of the outcome of these inspections.

3.2 Hypothesis formulation

The observations on the problem statement section lead us to testing four different basic hypotheses, to assess the effect of practitioners’ expertise on the outcome of the code inspection, in terms of the inspected defects diversity. We identify the hypothe-

13

ses as HA, HB, HC, and HD. For each of them, we formulate both a null and an alterna-tive hypothesis (e.g. HA0 and HA1).

As we shall see in the next section, we will break down each of these hypotheses into several specialized versions, to try out different expertise assessment metrics.

HA0: Developer skill has no effect on the inspected defect diversity. HA1: Developer skill has an effect on the inspected defect diversity. HB0: Peer skill has no effect on the inspected defect diversity. HB1: Peer skill has an effect on the inspected defect diversity. HC0: Reviewer expertise has no effect on the inspected defect diversity. HC1: Reviewer expertise has an effect on the inspected defect diversity. HD0: The gap of expertise between developer and peer has no effect on the in-

spected defect diversity. HD1: The gap of expertise between developer and peer has an effect on the in-

spected defect diversity.

3.3 Variables selection

Independent variables. The basic independent variable of this experiment is the sub-jects’ expertise. We use two measures of our subject’s expertise: their Average Grade (AG) throughout their academic path, based on the independent evaluation our sub-jects received in over 30 different courses, and the number of semesters (NSem) it took them to complete those courses. We assume that there is a higher merit in obtaining a given AG in the recommended number of semesters (RSem), than in a higher NSem. The Simple Weighted Average Grade (SWAG) and the Complex Weighted Average Grade (CWAG) expertise metrics, defined below, follow this rationale. Note that SWAG causes a bigger penalty than CWAG, as NSem increases.

),( NSemRSemMaxRSemAGSWAG ×=

),( NSemRSemMaxRSem

AGCWAG ×=

For each of the RTs, concerning expertise, we use the best subject, the worst sub-ject, and the average within the team. Finally, we also consider the difference between the expertise of the DT and that of the PT as an independent variable. In summary, we have 3 alternative rating schemes for grades, and 3 ways of combining grades within teams. This implies that we have 9 different ways for quantifying our independent variable (the expertise). These alternatives are used for each hypothesis under test.

Dependent variables. The dependent variables used in this experiment represent the diversity of defects found during code inspection. A defect classification checklist was distributed to all participants. The checklist contained 16 different defect classes, which were then subdivided into a total of 81 different defect codes. In summary, our dependent variables are:

• NDDClass: the number of different defect classes reported in the inspection; • NDDCode: the number of different defect codes reported in the inspection.

14

At first, NDDClass may seem unnecessary, given the usage of a finer grained meas-ure (NDDCode). However, if two code inspections report a similar number of different defect codes, but one of them uses a lot less defect classes than the other, it may be the case that this reflects a lower coverage of the kinds of problems to be found during the inspection. We used NDDClass to detect this kind of problem, should it occur.

3.4 Selection of subjects

The 87 subjects participating in this experiment are a convenient, but also representa-tive sample of the informatics students which annually graduate from our university (the numerus clausus for the informatics degree is 160, and the number of students graduating each year is around 60).

3.5 Experimental design

Regarding the experiment instrumentation, the calculation of subjects’ expertise was done upon the data available from the university’s academic database. The informa-tion concerning code inspections was collected from the normalized inspection reports submitted by subjects after they performed the Fagan inspections. Potential threats to validity [13], and how they were dealt with, are identified throughout the paper.

4 Data Collection

Preparation. The subjects were not aware of the aspects being researched, at the time they participated in the experiment, as this could jeopardize the validity of the results . They were only aware of our intention to use the data collected during the project.

Prior to the implementation of the components that were later inspected, subjects received a Java coding style guide, along with the set of standard public APIs for the components (specified as Java interfaces). Concerning code inspections, besides the referred checklist, subjects received a report template, so that they would perform the inspection and write down the report in a standardized fas hion. They also received training on how to perform Fagan inspections, prior to actually start ing them.

Execution. The experimental process was not allowed to disturb in any way the sub-jects’ activities in the projects . Subjects performed their normal tasks while developing this project, from requirements specification down to project delivery. Code inspection data was collected from the project’s deliverables, which was checked-in in a contents management system made available to students.

Data validation. In the beginning of the semester, there were 93 students enrolled in the course. Five of them dropped out befo re the experiment started, and one also gave up before turning in the first implementation of his group’s component. The remaining 87 students completed the project and are the subjects of this experiment. They were paired into 44 DTs . 43 of those DTs produced components that were inspected. The deliverables of these 43 inspections were used to collect the dependent variables.

15

5 Data analysis

We summarize the most relevant findings of our hypotheses tests . Further details are available at http://ctp.di.fct.unl.pt/QUASAR/Projects/CBSE/CodeInspections/.

5.1 Data set reduction

Outlier and extreme values can change our view on the relations between de-pendent and independent variables. For each dependent variable, we conducted a linear regression analysis using the average, best and worst cases of the independent variables. We repeated this analysis for each of our hypotheses and flagged as outliers those cases where the standard residual is greater than 1.5 times the standard deviation. This re-sulted in the removal of four cases in our analysis, with each of the depend-ent variables. The outlier removal proc-ess is illustrated in Fig. 2, where cases 15, 19, 25, and 38 are removed.

12,00 13,00 14,00 15,00 16,00 17,00

Peer Team Student's degree classification (best)

0

10

20

Nu

mb

er o

f d

iver

se s

pec

ific

bu

g c

od

esA

A

AA

AA

A

A

AA

AAA

A

A

A

A

A

A

A A

A

A

A

A

AA

A

A

A

A

A

A

AA

A

A

A

AA

AA

A

1

2

45

67

8

9

1011

121314

15

16

17

18

19

20

21 22

23

24

25

26

2728

29

30

31

32

33

34

3536

37

38

39

4142

4243

44

Fig. 2. Number of diverse specific bug codes, by PT expertise, including outliers.

5.2 Normality tests

We used the Kolmogorov-Smirnov (with the Lilliefors correction) to check normality. Using a confidence interval of 99% (test significance = 0.01), we can not reject the normality hypothesis, both for the independent and dependent variables. Therefore, we can use parame tric tests, such as the Pearson correlation coefficient.

5.3 Hypothesis testing

We started by performing correlation analysis, using the Pearson coefficient, among each of the independent and the dependent variables, to determine whether or not a relationship exists. These correlations were not significant with the predictors of hy-potheses HA and HC. As such, neither the influence of the DT skill, nor the influence of the overall RT skill in the outcome of the reviews, in terms of the diversity of the defect codes and classes , was confirmed.

From Table 1 we can observe significant correlations between our independent and dependent variables for hypotheses HB and HD. Most of the candidate predictors for HB have a significant positive correlation of above 40%. This relationship is observed both with NDDCode and NDDClass. The predictors for HD have a significant nega-tive correlation with the dependent variables. These correlations are stronger with SWAG and CWAG than with AG. Again, the same effect is observable both with NDDCode and NDDClass.

16

Table 1. Pearson correlations for the variables in hypotheses HB and HD, considering 39 cases (the outlier values referred in section 5.1 were removed, before the correlation analysis).

HB HD PT Diff_DT_PT AG sig. SWAG sig. CWAG sig. AG sig. SWAG sig. CWAG sig.

Avg. .469 .003 .429 .006 .462 .003 -.402 .011 -.454 .004 -.471 .002 Best .419 .008 .385 .016 .413 .009 -.409 .010 -.393 .013 -.419 .008

NDD Code

Worst .479 .002 .441 .005 .470 .003 -.350 .029 -.470 .003 -.468 .003 Avg. .416 .009 .378 .018 .407 .010 -.356 .026 -.433 .006 -.443 .005 Best .394 .013 .315 .051 .353 .028 -.376 .018 -.370 .020 -.395 .013

NDD Class

Worst .392 .013 .413 .009 .430 .006 -.293 .070 -.451 .004 -.451 .004

We further explored the HB and HD hypotheses, to check for significant differences observed in different groups of code reviews , using the ANOVA test. We started by computing the quartile values for the independent variables, and a ssigned the reviews to the respective quartile group. For each test, we had four groups with a growing expertise of the PT (in hypothesis HB), or with a growing difference between the exper-tise of the DT and the expertise of the PT (in hypothesis HD). The latter ranges from a negative value (DT has less expertise than PT) to a positive one (DT has a higher expertise than PT). Table 2 shows an example of this means comparison test, for hy-pothesis HD.

Concerning HB, we observed a variation among the different review groups that al-ways followed the same pattern. The reviews on the 4th quartile (the ones with the most expert peer teams) were always the ones with the highest NDDCode and NDDClass. Except when using predictors based on the worse PT element, or the aver-age value of SWAG, these differences were statistically signific ant. NDDCode and NDDClass showed an increase ranging from 36% to 111%. This trend is not visible in the first three quartiles, for some of the used metrics. The scatterplot presented on Fig. 2 is an example of a typical distribution of defect diversity vs. PT expertise. In sum-mary, we can reject the null hypothesis HB 0. We were able to find several measures of the expertise within the PT which can be used as predictors of the diversity of the reported defects .

With respect to HD, we observe the opposite pattern. With the expertise functions being used, the average number of different reported bug codes and classes decreased between 19% and 49%, when comparing the first with the last quartiles . In other words, the number of diverse defect codes and classes decreases as we move from DTs with lower expertise than their PTs to the opposite case. As such, we can reject the null hypothesis HD0. We found several measures of the difference between the expertise of the members of DT and PT which can be used as predictors of the diver-sity of the reported defects.

17

Table 2. Mean number of diverse defect codes found during code inspections. The difference between average AG of the DTs and PTs metric is used to place the PTs into the respective quartiles. Note that on the 1st quartile, DT has a much lower expertise than PT, while on the 4th, PT has a much lower expertise than DT.

Quartile Mean Diverse Defects N Std. Dev .

1st 7.00 10 2.828

2nd 5.40 10 3.688

3rd 6.10 10 2.558

4th 4.78 9 2.819

Total 5.85 39 3.005

6 Discussion

HA. We expected the best developers to produce components with fewer defects, but this was not confirmed. This result may be explainable in different ways. We did not use any information concerning neither the relative severity of the defects found in this analysis, nor their expected impact on maintenance. Moreover, we used defect code and class diversity, but not the actual number of reported defects in this analy-sis. Therefore, it may be the case that our dependent variable is too simplistic. It may also be the case that, because PTs were also part of the RTs, their expertise countered the effect of a lower variety of problems with that of a higher efficiency in finding them. A way to circumvent this would be to have several inspections being performed on the same artifacts by different teams , but this was not feasible in our context.

HB. As expected, we observed that the expertise of the PT does have a positive effect on the variety of problems uncovered during code inspections. We also note that the average and higher element expertise within the PT have stronger correlations with the outcome of the review than the expertise of the “weaker” element of the PT. Along with the significant boost of results with the PTs on the best quartile, this increases our confidence on the positive effect of expert peer reviewers in the reviewer team and also points to a small effect of “leadership” within those teams.

HC. The expertise of the whole review team did not show a significant relationship with the outcome of the review. The considerations concerning a possible over-simplification of our dependent variable, combined with the cancellation effect also described with respect to HA may be responsible for this discrepancy between the expected result and the outcome of this experiment.

HD. As expected, when PTs of low expertise analyze the work of DTs with a higher expertise, the outcome of the code review shows a lower variety of defects found. Conversely, more defects are found in inspections where the PTs have a higher exper-tise than the one of the DTs. A potential leadership effect of a reviewer over the others is not visible from the data analyzed while testing this hypothesis.

With the experiment design of this last hypothesis , we have an alternative perspec-tive on the inspection group dynamics, when compared to hypothesis HC. On HC we had no indication of how the expertise was distributed within the group, thus being

18

vulnerable to the cancellat ion effect occurring when (i) having good experts examining their own code and not finding many problems with it, because they were not there, or (ii) weaker programmers examining their own code and not realizing the problems in it. Both situations lead to a cancellation effect that might explain the unexpected results with hypothesis HC.

There is a curious effect in the evolution of the variety of defects found between the second and third quartiles of HD (the second quartile has DTs with a lower exper-tise than their PTs, while the third inverts this relationship). One could expect the variety of defects to be lower on the third quartile, following the tendency found from the first to the forth quartiles. However, the expertise level is very close, within groups 2 and 3. Therefore, it may be the case that it is the domain level expertise that domi-nates the outcome of the inspection. With a better knowledge of the deliverables b e-ing inspected, allied with a slightly better expertise than their peers, the authors may be responsible for this locally increased benefit of the code review. As the gap of expertise between DT and PT members widens, this effect would be mitigated by the dominating effect of the higher code quality and lower external reviewer expertise.

7 Conclusions

We described an exp eriment carried out to help understanding the effect of practitio-ner’s expertise in the deliverables produced in the context of CB development.

We focused our attention on the outcome of code inspections, and, in particular, on the variety of problems reported during those inspections. We confirmed the expected positive effect of the expertise of the peer review teams in the outcome of the reviews, observable through the increased variety of defects found when peer experts were available. We also confirmed that having expert peers collaborating in the inspection of components developed by less skilled peers has a positive impact on the outcome of the review. Moreover, there is also a learning effect, not studied here but vastly commented on the literature, when combining experts with non-experts. This is also expected for the opposite case, where non-experts participate on the review of code developed by experts. However, in this case, a lower variety of defects is found, both because the code is likely to have a higher quality, and because the external reviewers have less capacity to detect its problems. Given the main goal of inspections (maximiz-ing defect detection), the results are poorer.

When observed in isolation, the expertise of the DTs did not show a significant re-lationship with the variety of problems found. The expertise of the RTs was also not shown to be a good indicator of the outcome of the inspection. Further research is required to determine whether these were the results of cancellation effects of exper-tise, or if more sophisticated review outcome metrics should have been used here.

As future work, we expect to expand on this experiment by exploring this interpreta-tion of why two of our hypotheses were not confirmed. The deliverables of the project that served as a basis for this experiment include some details that were not explored in this paper, such as code complexity metrics, and the practitioners’ assessment of the potential impact of the problems reported. We plan to further explore these data to

19

strengthen the conclusions reported here and to explore other related hypotheses on the effect of expertise throughout the development process.

Acknowledgments

This work is sponsored by the FCT STACOS project (POSI/CHS/48875/2002).

References

1. Fagan, M.E., Design and Code Inspections to Reduce Errors in Program Development. IBM Systems Journal, 1976. 15(3): p. 182-211.

2. Fagan, M.E., Advances in Software Inspections. IEEE Transactions on Software Engineer-ing., 1986. 12(7): p. 744-753.

3. Parnas, D.L. and Weiss, D.M., Active Design Reviews: Principles and Practices. Journal of Systems and Software, 1987. 4(7): p. 259-265.

4. Knight, J.C. and Myers, E.A., An Improved Inspection Technique. Communications of the ACM, 1993. 36(11): p. 51-61.

5. Porter, A. and Votta, L., What Makes Inspections Work? IEEE Software, 1997. 14(6): p. 99-102.

6. Weller, E.F., Lessons from Three Years of Inspection Data. IEEE Software, 1993. 10(5): p. 38-45.

7. Siy, H., Identifying the Mechanisms to Improve Code Inspection Costs and Benefits , PhD Thesis, University of Maryland, USA. 1996.

8. Porter, A., Siy, H., Mockus, A., and Votta, L., Understanding the sources of variation in software inspections. ACM Transactions on Software Engineering and Methodology, 1998. 7(1): p. 41-79.

9. Biffl, S. and Halling, M. Investigating the Influence of Inspector Capability Factors with Four Inspection Techniques on Inspection Performance . in Eighth IEEE International Sym-posium on Software Metrics (Metrics'02). 2002.

10. Laitenberger, O. and DeBaud, J.-M., An Encompassing Life-Cycle Centric Survey on Soft-ware Inspection. Journal for Systems and Software, 2000. 50(1): p. 5-31.

11. Sauer, C., Jeffery, R., Land, L., and Yetton, P., The Effectiveness of Software Development Technical Reviews: A Behaviorally Motivated Program of Research. IEEE Transactions on Software Engineering, 2000. 26(1): p. 1-14.

12. Kruger, J. and Dunning, D., Unskilled and Unaware of It: How Difficulties in Recognizing One's Own Incompetence Lead to Inflated Self-Assessments. Journal of Personality and So-cial Psycology, 1999. 77(6): p. 1121-1134.

13. Wohlin, C., Runeson, P., Höst, M., Ohlsson, M.C., Regnell, B., and Wesslén, A., Experi-mentation in Software Engineering: An Introduction . Vol. 6. 1999, Boston, EUA: Kluwer Academic Publishers. 224 pages.

20

A Substitution Model for Software Components

Bart George, Régis Fleurquin, and Salah Sadou

VALORIA Lab., University of South Brittany, France{Bart.George,Regis.Fleurquin,Salah.Sadou}@univ-ubs.fr

Abstract. One of Software Engineering’s main goals is to build com-plex applications in a simple way. For that, software components mustbe described by its functional and non-functional properties. Then, theproblem is to know which component satisfies a specific need in a spe-cific composition context, during software conception or maintenance.We state that this is a substitution problem in any of the two cases.From this statement, we propose a need-aware substitution model thattakes into account functional and non-functional properties.

1 Introduction

Component-oriented programming should allow us to build a software like apuzzle whose parts would be units ”subjects to composition by a third party”[13]. Examples of such units are COTS (Components-Off-The-Shelf ), which arecommercial products from several constructors and origins. When one developsand maintains a component-based software, some problems occur, and we willnotice two main ones: how to select, during conception of such a software, themost suitable component in order to satisfy an identified need ? And during amaintenance, if this need evolves, will the chosen component remain suitable, orshall we replace it ?

We think that these problems are related to a substitution problem. In fact,when one conceives or maintains an application, some needs appear. And todescribe them, the designer or the maintainer can imagine ideal components.These are virtual components representing the best ones satisfying these specificneeds. Then the problem is to find the concrete components which are the closestto the ideal ones. In other words, trying to compose or maintain componentsmeans trying to make concrete components substitute ideal ones.

However, composition doesn’t concern only the functional aspect. Most com-ponents are ”black boxes” which must describe not only functional, but also non-functional properties. As every software needs a certain quality, one can’t thinkabout composing components whose non-functional properties are unknown, andat the same time hope having its quality requirements satisfied anyway. This iswhy substitution must take functional and non-functional properties into ac-count.

So, how to substitute ? Some may say we just have to use subtyping, as someobject-oriented languages made it a general way of substitution. However, anideal component describes more than general needs: it describes the application’s

21

context, a notion that is absent from objects. Let us explain what we mean by”context”. If we take a need, modeled by an ideal component, we will try to find aconcrete one to substitute it. Now, let us suppose that we already found a suitablecomponent. We may need to check if there isn’t another one better than the firstone. However, trying to substitute the old candidate by a new one would be amistake, because the key notion isn’t the candidate, but the need it is supposedto satisfy. Plus, if this need changes, a former candidate may no longer remainsuitable. So substitution of an ideal component by a concrete one is performedonly into the context of the need modeled by the ideal component. This is whya candidate component can replace another one without any subtyping relationbetween them, as every candidate is compared only to the ideal component.

In this paper, we consider a generic component model and a quality model,and into this framework we define (section 2) a component-oriented substitutionmodel, including a distance from a candidate component to an ideal one. Inorder to illustrate the possibilities of such a model, we describe the differentsubstitution cases during the life cycle using a short application example (section3). Then, before concluding, we describe some related works (section 4).

2 Our substitution model

Definitions given in this paper are placed in the following framework: one compo-nent model, holding a type system such as Java for EJB, and one quality modelsuch as ISO 9126 standard [11]. In this framework, we suppose the existence ofmetrics to measure non-functional properties, so that our contribution will focusonly on the substitution model definition.

Here, we will present only the basic concepts of this model. A more detaileddescription is available in [7]. Note that in this version of our work, we performsubstitution at the individual component level.

2.1 Component and quality generic models

Our goal is not to give yet another definition of what a component is, or whatnon-functional properties are. It is to define a component-oriented substitutionthat we can apply on many existing component and quality models. That iswhy we prefer to give generic models, on which we can apply our substitutionconcepts.

The generic component model includes component artifacts, representingthe component’s architectural elements, which are common to most existing com-ponent models, and which have non-functional properties. As shown in figure 1,we chose to keep three kinds of component artifacts: components themselves, in-terfaces, and operations. A component contains provided and required interfaces,and interfaces contain operations. In the remaining of the paper, we refer to can-didate component and substitutable component when the first one tries tosubstitute the second one. Their elements are called respectively candidate el-ements and substitutable elements. When we find the best candidate for the

22

Fig. 1. Our generic model

substitution, we say the substitutable component or element can be replacedby this candidate.

Beside the component model, we define a generic quality model, on whichthe quality properties of the component model’s elements are based. Elements ofthe quality model are quality characteristics (such as those from ISO 9126 [11]),and metrics. We use existing metrics to evaluate and compare non-functionalproperties (see [8] for a survey). But why metrics ? In the literature, severalmethods for defining and evaluating non-functional properties already exist (see[1] for a survey). But such methods usually focus on one specific property, orfamily of properties, for example quality of service, which is only a part of thewhole software quality. Metrics may be applied to many families of properties,and allow comparisons. This is why we think that in our case, metrics representthe best method for comparing different non-functional properties.

2.2 Non-functional specifications

Elements of the component model are linked to elements of the quality modelusing a non-functional specification (noted NFS). An artifact may be relatedto several quality elements, so several NFSs belong to only one artifact. AnNFS describes the effect of a quality characteristic on the artifact it belongs to,and uses the metric applied on the latter. Several NFSs of a same componentartifact may share the same metric, but not the same characteristic. The set ofan artifact’s NFSs is called a quality field.

In Figure 1, the resultValue attribute of an NFS is given by the metric’smeasurement on the artifact. In the case of an ideal component, this attributevalue is given by the application’s designer.

23

2.3 Comparability of elements

We can try to compare two NFSs only if we can compare the artifacts theybelong to. And we can try to compare artifacts of the same kind only. TwoNFSs of comparable artifacts are comparable only if they measure the samecharacteristic (which means they use the same metric too, as one characteristicis measured by only one metric). Two NFSs are equal if they are comparableand their resultValue attributes are equal.

Two operations are comparable if their signatures are comparable. Two op-erations are equal if their signatures are equal modulo the renaming of the typenames, and if their quality fields are equal.

A candidate provided interface PI1 is comparable to a substitutable providedinterface PI0 if for each operation of PI0 there exists a comparable operationin PI1. A candidate required interface RI1 is comparable to a substitutablerequired interface RI0 if for each operation of RI1, there exists a comparableoperat