Re: git: test suite fails on hppa, in run_with_limited_stack
On Wed, May 13, 2020 at 4:53 AM John David Anglin <dave.anglin@bell.net> wrote:
> On 2020-05-13 1:44 a.m., Greg Price wrote:
> > Probably cleanest to put this outside the `case $uname_s`.
> I put it where I did because the result of uname -m is somewhat OS dependent.
>
> The config.guess script attempts to guess a canonical system name for a target. It has all the gory
> details on the interpretation of uname -m.
I just took a look through that. Wow, I see what you mean!
case "$UNAME_MACHINE:$UNAME_SYSTEM:$UNAME_RELEASE:$UNAME_VERSION" in
# ...
9000/7??:4.3bsd:*:* | 9000/8?[79]:4.3bsd:*:*)
echo hppa1.1-hp-bsd
exit ;;
And there are lots more like that one, and others with complex internal logic.
Still I'm not sure that that situation is helped by putting the `uname
-m` inside this case. This is the catchall `*)` case after looking
only for MINGW, CYGWIN, and FreeBSD, and those all appear to have
pretty sane `uname -m` behavior. (On FreeBSD, `uname -p` gets a bit
more specific but `uname -m` is already sane; I found the report here:
https://github.com/mesonbuild/meson/issues/5766
informative.) If any of the OSes where config.guess has to do more
work were to get this far in building and testing Git, they'd end up
running any code we stick at this spot anyway. So I think it still
makes more sense to put the new `case` next to the existing one.
One other thought I have from that config.guess example, though --
there are some cases there like this:
parisc64:Linux:*:* | hppa64:Linux:*:*)
echo hppa64-unknown-linux-"$LIBC"
exit ;;
That suggests to me that the condition should perhaps look for "hppa", too, like
parisc* | hppa*)
test_set_prereq HPPA
;;
What do you think?
Cheers,
Greg
Reply to: