[Haskell-cafe] How to diagnose a loop in a large Haskell program?

[Han Joosten]

Thanks for the pointers. I never realized that "simulated call-stack" could
be something different than the HasCallStack stuff. When I used the
following command, I get the callstack info I was looking for:

stack install --profile --no-strip --ghc-options -fprof-auto-calls &&
ampersand -- check
testing/Sentinel/Tests/ShouldSucceed/OnlyValidation/Issue280.adl

I hope this will help me to find the loop.

Op do 22 sep. 2022 om 11:52 schreef Dominik Schrempf <
dominik.schrempf at gmail.com>:

> I just noticed, maybe you are interested in the `-xc` runtime system
> option? This option can be activated when compiling with profiling
> support. See here:
>
>
> https://downloads.haskell.org/ghc/latest/docs/users_guide/profiling.html#rts-flag--xc
>
> Dominik
>
> Dominik Schrempf <dominik.schrempf at gmail.com> writes:
>
> > Han Joosten <han.joosten.han at gmail.com> writes:
> >
> >> Hi,
> >>
> >> I am struggling to debug a loop (some nontermination behaviour) of a
> huge program. Normally, I use `stack build` to generate it. I have tried to
> use
> >> `trace` at lots of places in the code, recompile and see what happens.
> However, it isn't always clear in what order the trace statements are fired,
> >> because of the lazy evaluation of Haskell.
> >>
> >> So I tried another approach. I have a function
> >>
> >> fatal :: HasCallStack => Text -> a
> >> fatal msg =
> >>   exitWith . Fatal . T.lines $
> >>     ( msg <> "\n"
> >>         <> (utf8BuilderToText . displayCallStackFull $ callStack)
> >>     )
> >>
> >> displayCallStackFull :: CallStack -> Utf8Builder
> >> displayCallStackFull cs =
> >>   case reverse $ getCallStack cs of
> >>     [] -> "<no call stack found>"
> >>     xs -> mconcat $ fmap showCall xs
> >>   where
> >>     showCall :: (String, SrcLoc) -> Utf8Builder
> >>     showCall (desc, loc) =
> >>       let file = srcLocFile loc
> >>        in "\n"
> >>             <> fromString file
> >>             <> ":"
> >>             <> displayShow (srcLocStartLine loc)
> >>             <> ":"
> >>             <> displayShow (srcLocStartCol loc)
> >>             <> " "
> >>             <> fromString desc
> >>
> >> (The function to show the stack was inspired from
> https://hackage.haskell.org/package/rio-0.1.22.0/docs/RIO.html#v:displayCallStack
> ).
> >>
> >> I would expect that at any place where I force a call to `fatal`, I
> would get the full stacktrace. Unfortunately, I only get the place where
> the call to
> >> fatal is in the code.
> >
> > Hi.
> >
> > Do the calling functions have call stack constraints?
> >
> > Quoting from the GHC manual:
> https://ghc.gitlab.haskell.org/ghc/doc/users_guide/exts/callstack.html
> >
> > "The call stack will only extend as far as the types allow it."
> >
> > I think compilation with `--profile` (stack) or `--enable-profiling`
> (cabal-install) activates "simulated call-stacks" which seem to differ from
> the call stacks obtained by `HasCallStack`. Is this correct?
> >
> > On a side note: I usually get a complete call stacks when compiling with
> `cabal build --enable-profiling`.
> >
> > Dominik
> >
> >>
> >> I have been looking at the documentation of both stack and ghc, but I
> don't understand how to compile my code in a way that the full stacktrace is
> >> shown.
> >>
> >> Until now the best guess I have is
> >>
> >> stack install --profile --no-strip && ampersand -- check
> testing/Sentinel/Tests/ShouldSucceed/OnlyValidation/Issue280.adl
> >>
> >> The ultimate goal is to get rid of this ugly bug regarding the loop.
> Any help/suggestions is really appreciated!
> >>
> >> Thanks for reading.
> >> Han Joosten
> >>
> >> _______________________________________________
> >> Haskell-Cafe mailing list
> >> To (un)subscribe, modify options or view archives go to:
> >> http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe
> >> Only members subscribed via the mailman list are allowed to post.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/haskell-cafe/attachments/20220922/32794e13/attachment.html>