On 08.04.2015, at 20:29, Eliot Miranda <eliot.miranda at gmail.com> wrote:

> On Wed, Apr 8, 2015 at 11:11 AM, Eliot Miranda <eliot.miranda at gmail.com>
> wrote:
> 
> > On Tue, Apr 7, 2015 at 7:39 AM, Levente Uzonyi <leves at elte.hu> wrote:
> >
> >> The decompiler is not perfect, and IIRC someone stated that it will never
> >> be. Most of these errors are "handled" by adding them as exceptions, so
> >> they are not tested. Some of these issues are probably fixable, but not
> >> many people are familiar with the decompiler.
> >> See DecompilerTests >> #decompilerFailures for the list of currently
> >> ignored methods.
> >>
> >> Levente
> >>
> >>
> >> On Tue, 7 Apr 2015, Tobias Pape wrote:
> >>
> >>  Hi
> >>>
> >>>
> >>> On 04.04.2015, at 16:59, Frank Shearar <frank.shearar at gmail.com> wrote:
> >>>
> >>>  On 3 April 2015 at 16:27, Tobias Pape <Das.Linux at gmx.de> wrote:
> >>>>
> >>>>> Hey
> >>>>>
> >>>>> I am watching the CI doing its work as changes and fixes
> >>>>> chime in.
> >>>>>  I noticed that there are still 13 test failing[1] and I'm
> >>>>> not sure how all of them should be fixed :)
> >>>>>  Let's make it green!
> >>>>>
> >>>>
> >>>> I am _very very happy_ to see someone other than me harping on about
> >>>> tests! Please carry on!
> >>>>
> >>>> This one ought to be easy to fix: it's a small breakage in the
> >>>> modularity of the image; some methods need to be recategorised as not
> >>>> being System: http://build.squeak.org/job/SqueakTrunk/1207/testReport/
> >>>> Tests.Dependencies/PackageDependencyTest/testSystem/
> >>>>
> >>>> This one too: http://build.squeak.org/job/SqueakTrunk/1207/testReport/
> >>>> Tests.Dependencies/PackageDependencyTest/testTools/
> >>>> (Tools shouldn't depend on _ToolBuilder-Morphic_, even though it would
> >>>> be OK to depend on ToolBuilder.)
> >>>>
> >>>> And I love that this one fails -
> >>>> http://build.squeak.org/job/SqueakTrunk/1207/testReport/
> >>>> Tests.Dependencies/PackageDependencyTest/testMultilingual/
> >>>> - Multilingual no longer depends on TrueType, and our system just got
> >>>> one dependency cleaner.
> >>>>
> >>>
> >>> I did some things :)
> >>>
> >>> But I need help:
> >>> There's a strange thing in the decompiler/compiler:
> >>>
> >>> http://build.squeak.org/job/SqueakTrunk/lastCompletedBuild/testReport/
> >>> Tests.Compiler/DecompilerTests/testDecompilerInClassesBAtoBM/
> >>>
> >>> Somewhere in Behavior>>#toolIconSelector:,
> >>> a sequence of 'push nil. pop' is generated that is decompiled as 'nil.'.
> >>> This does not match the source and I don't know whether it's a bug in
> >>> the Compiler emitting this or the Decompiler not recognizing the emitted
> >>> pattern.
> >>>
> >>
> > Ah, I see it.
> >
> > Look the end of this block:
> > self methodDictionary at: aSymbol ifPresent: [ :method |
> > ...
> > method hasReportableSlip ifTrue: [^ #breakpoint]].
> >
> > The block (correctly) answers nil if method hasReportableSlip is false,
> > and the bytecode contains a pushConstant: nil; blockReturn sequence to
> > implement this.  So the problem is with the decompiler not eliminating the
> > explicit nil node in this case where it should be implicit.
> >
> 
> Oops.  The pushConstant: nil; blockReturn sequence is at bytecpde pc 202
> right? That's coming from
> 
> (self isSelectorOverride: aSymbol)
> ifTrue: [
> (self isSelectorOverridden: aSymbol)
> ifTrue: [ ^ #arrowUpAndDown ]
> ifFalse: [ ^ #arrowUp ] ]
> ifFalse: [
> (self isSelectorOverridden: aSymbol)
> ifTrue: [^ #arrowDown ]].
> 

Yes, this is the place :)

> I'll try and squash this in a principled way soon.  I've squashed the
> former by using the attached:
> 

Thanks for taking care.
Best
	-Tobias

> 
> 
> -- 
> best,
> Eliot