Re: Bug in mk/plist/plist-info.awk

[Aleksej Lebedev <root%zta.lk@localhost>]

On 2019-02-13 05:01, David H. Gutteridge wrote:
> On Mon, 11 Feb 2019, at 17:28:02 +0100, Aleksej Lebedev wrote:
> > Hi.
> >
> > I found a bug in mk/plist/plist-info.awk, which I am hesitating to fix
> > myself because I don't fully understand why the following reg-ex is
> > written like it is.
> >
> >
> > It's the line 103:
> >
> >     /^([^\/]*\/)*(info\/[^\/]+(\.info)?|[^\/]+\.info)$/ {
> >
> > Because of this ? after (\.info) some unrelated lines like this one:
> >
> > lib/python3.6/site-
> > packages/conans/test/unittests/model/info/__init__.py
> >
> >
> > get (incorrectly) matched which results in them being skipped, while
> > when PLIST is processed by lang/python/plist-python.awk later.
> >
> > I am not 100% if this ? was added by mistake, but it results in files
> > like
> >
> > one mentioned above being matched.
> >
> > Not sure I understand why the cases
> >
> >     path/to/some/info/file.info
> >
> > and
> >
> >     path/to/some/file.info
> >
> > are matched separately.
> >
> > So to me, the whole reg-ex should be written as:
> >
> >     /^([^\/]*\/)*[^\/]+\.info$/
> >
> > But I might miss something.
>
> There is a workaround you can apply in your package:
>
> diff --git a/py-conan/Makefile b/py-conan/Makefile
> index 6a48049017..9a005d6a7a 100644
> --- a/py-conan/Makefile
> +++ b/py-conan/Makefile
> @@ -12,6 +12,7 @@ LICENSE=      mit
>
>  USE_LANGUAGES= # none
>
> +IGNORE_INFO_DIRS+=     ${PYSITELIB}/conans/test/unittests/model/info
>
>  post-install:
>         cd ${DESTDIR}${PREFIX}/bin && for b in conan conan_build_info
> conan_server; \
>
> (Though I still argue that approach should be opt-in for packages that
> need to explicitly deal with info files missing the expected extension,
> not opt-out for packages that don't need it.)
>
> Dave

Thank you. I didn't think about IGNORE_INFO_DIRS. I agree with you about
what should be opt-in & opt-out.

Meanwhile I was trying to come up with the proper fix (if there is a 
proper fix).
But I am busy with other thinks a bit.

In any case, what I did is I searched for all PLISTS
that contain some files that match the current regex. It's huge -- 204 
including wip
and 84 not including wip. Considering the fact that the fix was done to 
make
some (a few) packages work (in the ticket they mention "packages (such
as emacs)"), this fix is way too inappropriate in the sense that any of 
those 204
packages might be broken any time they require some special treatment 
with PLIST_VARS.

On the other hand I wander if we really need to have this "next;" 
statement
at the end of the last cause in mk/plist/plist-info.awk?

For example, lang/python/plist-python.awk doesn't cancel other awk 
clauses after
handling entries with [py3x], [py2x].

If we remove this "next" statement then all the files false-positively 
matched as info files
will still be processed with further awk clauses.

I will test it a bit and try to come up with a patch.

--
Aleksej Lebedev