On 21/03/2023 18:08, George Valkov wrote:
On 2023-03-21, at 7:32 PM, Paul Eggert <egg...@cs.ucla.edu> wrote:
On 3/21/23 08:27, George Valkov wrote:
However since the old m4-1.4.19 in under tools/m4, coreutils builds with the
legacy version:
If you're building from that tarball, coreutils doesn't need m4. I just now double-checked that,
in an environment that didn't have m4. "./configure && make check" succeeded.
So the m4 version should be irrelevant.
Yes, I can confirm that a stand-alone build using your instructions works. I do
not have enough experience with the OpenWRT build system, but what I can see is
that at some point it tries to run staging_dir/host/bin/m4
And that’s why I had the impression that the legacy version of m4 is used. I
might be wrong, and maybe it just needs that executable, since when tools/m4 is
present, tools/coreutils builds and works. If the m4 directory from the tarball
is used, then everything’s fine, we can keep tools/m4 as it is.
staging_dir/host/bin/m4 is compiled from tools/m4 and does not exist without
tools/m4, so the build fails if I remove it. If I create a link to
/usr/local/Cellar/m4/1.4.19/bin/m4, that part of the build succeeds, but then
we also need to apply the following patch:
https://github.com/openwrt/openwrt/blob/master/tools/coreutils/patches/001-m4.patch
Do you know if the patch is applied upstream in coreutils?
Yes if that m4 file is patched then m4 will need to run to regen stuff.
Though the existing m4 tool should be fine for that.
I would look at figuring out why that m4 tool is not available in your env.
Perhaps you need to define an M4 env variable or something.
The openwrt m4 patch originates from
https://git.openwrt.org/?p=openwrt/openwrt.git;a=commitdiff;h=c5608227
and that pattern seems to have originated from:
https://github.com/openwrt/packages/issues/14120#issuecomment-749438742
For the patch not to be needed it would need to be applied to gnulib,
but I don't know how general it is.
cheers,
Pádraig
