~rattab/autoconf-upstream

9276ff1bac7072e03900283cd76e23226656cf6d — Zack Weinberg 1 year, 8 months ago 8a7bda3
Review OpenEmbedded/Yocto patch list.
1 files changed, 41 insertions(+), 2 deletions(-)

M todo.org
M todo.org => todo.org +41 -2
@@ 222,7 222,17 @@ Upstream status of the patch is "Innapropriate [configuration]"

Seems to be tweaks to the build to enable `--warning=cross` if available.

** QUESTION oe/config_site.patch
[ZW: This is working around a bug in all currently-released versions
of Automake: `--warning=cross` is only understood by `autoconf`, not
`automake`.  (The set of valid arguments to the `--warning=` option is
_supposed_ to be consistent across all of the `auto*` programs, so
that autoreconf doesn’t have to care about this sort of thing.)
Unfortunately the bug is hard to fix at source, due to the awkward way
some library files are shared between Autoconf and Automake.  There
is an argument for taking this patch.  I will start a discussion on
the mailing lists.]

** SKIP oe/config_site.patch
The patch is supposed to allow for multiple files to be provided in CONFIG_SITE
but it doesn't change the processing of the variable at all. Instead it seems
like it's removing the default search paths for the config.site file.


@@ 230,22 240,39 @@ like it's removing the default search paths for the config.site file.
Comment on the patch seems to indicate that in 2.68 the variable wasn't being
used but nothing of what I'm reading in the patch seems to agree with this.

[ZW: I see what this is trying to do, but it’s not correct as is.
It loses the check for both $prefix/share/config.site and
$prefix/etc/config.site when $CONFIG_SITE *isn’t* set.
Post-2.70 material in any case.  Please file a bug for this one
so we remember to come back to it.]

** SKIP oe/fix_path_xtra.patch
This patch just removes XMKMF to remove confusion in their particular build
enviornment. I'm going to go ahead and say we can skip this.

[ZW: It sounds like this patch is motivated by a real problem with
Autoconf’s support for cross-compiling X11 programs, but we definitely
wouldn’t take it as-is, and I think a proper fix would be complicated
enough that it should wait for 2.71.  Please file a bug for this one
so we remember to come back to it.]

** SKIP oe/performance.patch
2b59b6f8a79b8bf77e178ff4e5aa0ede433d39cf

From a quick reading, it seems like the above patch does essentially the same
thing but in a cleaner way.

[ZW: Yep, 2b59b6f8a79b8bf77e178ff4e5aa0ede433d39cf does the same thing
but more thoroughly.]

** SKIP oe/preferbash.patch
Prefers bash over sh because in their system sh can map to dash and they use a
lot of bashisms.

Highly specific to Yocto so shouldn't be upstreamed.

[ZW: Concur.]

** EASY oe/program_prefix.patch
Only description for the patch is a dead link.



@@ 254,6 281,9 @@ Adds a check for `--target != --host` prior to setting program_prefix to

I can't tell whether this is a good idea or not.

[ZW: Please file a bug for this one.  We need to ask the OE people for
a rationale, but it *might* be appropriate to take for 2.70.]

** QUESTION oe/remove-usr-local-lib-from-m4.patch
Looks like autoconf uses a hard coded coded /usr/local path for AIX which causes
yocto some problems (these problems are not described).


@@ 262,7 292,16 @@ The patch itself just removes a problematic section for AIX which they don't
support. Could investigate whether replacing the hard-coded path by something
else could solve their problem.

* TODO openbsd
[ZW: It might be possible to write this AIX-specific logic in a
different way, that would avoid having /usr/local/lib searched for
libraries other than the bolt-on getloadavg implementation that’s
wanted here.  However, to code it up, we would need to find someone
with access to an AIX host *that has this nonstandard library
installed*.  I have access to AIX but not with this library.  Until we
can find someone I think we should leave well enough alone.  However,
please file a bug so we don’t forget about it.]

* QUESTION openbsd
Gave up trying to extract information out of the commit history because the
folder structure complicates things quite a bit and the commit messages don't
add any useful context. Need to hunt down mailing list threads for additional