[BUG] Re: Git 2.35.0-rc0

> Hi All,
>
> Git 2.35.0-rc0 is not portable:
>
> uncompress2(uncompressed + block_header_skip, &dst_len,
>                    ^
> "/home/ituglib/randall/jenkins/.jenkins/workspace/Git_Pipeline/reftable/bloc
> k.c", line 213: error(114):
>           identifier "uncompress2" is undefined
>
> This is not a POSIX compliant call.

> On Mon, Jan 10, 2022 at 09:20:15PM -0500, rsbecker@nexbridge.com
> wrote:
> > Hi All,
> >
> > Git 2.35.0-rc0 is not portable:
> >
> > uncompress2(uncompressed + block_header_skip, &dst_len,
> >                    ^
> > "/home/ituglib/randall/jenkins/.jenkins/workspace/Git_Pipeline/reftabl
> > e/bloc
> > k.c", line 213: error(114):
> >           identifier "uncompress2" is undefined
> >
> > This is not a POSIX compliant call.
> 
> uncompress2() isn't a POSIX thing, it's part of zlib IIUC. The new caller is in the
> reftable code, which provides a build knob
> NO_UNCOMPRESS2 that you can use if your version of zlib does not have
> uncompress2().
> 
> uncompress2() was added to zlib back in 37281ac (Add uncompress2()
> function, which returns the input size used., 2016-11-15), which first
> appeared in version 1.2.9.
> 
> If your system doesn't have a modern-ish zlib, you may try building with that
> knob, or upgrading your system's copy of zlib. And if NonStop doesn't have a
> modern zlib available at all, we should modify the NonStop section of
> config.mak.uname.

> > If your system doesn't have a modern-ish zlib, you may try building with that
> > knob, or upgrading your system's copy of zlib. And if NonStop doesn't have a
> > modern zlib available at all, we should modify the NonStop section of
> > config.mak.uname.
>
> There is no provision in reftable/block.c to avoid using uncompress2,
> so the knob will not help. Our zlibc is not that recent (as in it does
> not have uncompress2) and we cannot make the 2.35.0 timeframe to
> upgrade it. The current zlib seems to require gcc and is very
> difficult to port at this stage. This is a blocker situation.

> To: rsbecker@nexbridge.com
> On Mon, Jan 10, 2022 at 09:57:57PM -0500, rsbecker@nexbridge.com
> wrote:
> > > If your system doesn't have a modern-ish zlib, you may try building
> > > with that knob, or upgrading your system's copy of zlib. And if
> > > NonStop doesn't have a modern zlib available at all, we should
> > > modify the NonStop section of config.mak.uname.
> >
> > There is no provision in reftable/block.c to avoid using uncompress2,
> > so the knob will not help. Our zlibc is not that recent (as in it does
> > not have uncompress2) and we cannot make the 2.35.0 timeframe to
> > upgrade it. The current zlib seems to require gcc and is very
> > difficult to port at this stage. This is a blocker situation.
> 
> NO_UNCOMPRESS2 does not avoid calling uncompress2, but instead compiles
> a copy-and-pasted implementation in compat/ so that the function is available.
> 
> Looking through it, I can't imagine that it wouldn't compile just fine even on
> NonStop.
> 
> Have you tried building with NO_UNCOMPRESS2?

> On Mon, Jan 10, 2022 at 09:57:57PM -0500, rsbecker@nexbridge.com
> wrote:
> > > If your system doesn't have a modern-ish zlib, you may try building
> > > with that knob, or upgrading your system's copy of zlib. And if
> > > NonStop doesn't have a modern zlib available at all, we should
> > > modify the NonStop section of config.mak.uname.
> >
> > There is no provision in reftable/block.c to avoid using uncompress2,
> > so the knob will not help. Our zlibc is not that recent (as in it does
> > not have uncompress2) and we cannot make the 2.35.0 timeframe to
> > upgrade it. The current zlib seems to require gcc and is very
> > difficult to port at this stage. This is a blocker situation.
> 
> NO_UNCOMPRESS2 does not avoid calling uncompress2, but instead compiles
> a copy-and-pasted implementation in compat/ so that the function is available.
> 
> Looking through it, I can't imagine that it wouldn't compile just fine even on
> NonStop.
> 
> Have you tried building with NO_UNCOMPRESS2?

> On January 10, 2022 10:00 PM, Taylor Blau wrote:
> > On Mon, Jan 10, 2022 at 09:57:57PM -0500, rsbecker@nexbridge.com
> > wrote:
> > > > If your system doesn't have a modern-ish zlib, you may try
> > > > building with that knob, or upgrading your system's copy of zlib.
> > > > And if NonStop doesn't have a modern zlib available at all, we
> > > > should modify the NonStop section of config.mak.uname.
> > >
> > > There is no provision in reftable/block.c to avoid using
> > > uncompress2, so the knob will not help. Our zlibc is not that recent
> > > (as in it does not have uncompress2) and we cannot make the 2.35.0
> > > timeframe to upgrade it. The current zlib seems to require gcc and
> > > is very difficult to port at this stage. This is a blocker situation.
> >
> > NO_UNCOMPRESS2 does not avoid calling uncompress2, but instead
> > compiles a copy-and-pasted implementation in compat/ so that the function
> is available.
> >
> > Looking through it, I can't imagine that it wouldn't compile just fine
> > even on NonStop.
> >
> > Have you tried building with NO_UNCOMPRESS2?
> 
> The patch for NonStop to make this compile is as follows. Test is running - will
> be 2 days:
> 
> diff --git a/config.mak.uname b/config.mak.uname index
> a3a779327f..9b3e9bff5f 100644
> --- a/config.mak.uname
> +++ b/config.mak.uname
> @@ -576,6 +576,7 @@ ifeq ($(uname_S),NONSTOP_KERNEL)
>         NO_SETENV = YesPlease
>         NO_UNSETENV = YesPlease
>         NO_MKDTEMP = YesPlease
> +       NO_UNCOMPRESS2 = YesPlease
>         # Currently libiconv-1.9.1.
>         OLD_ICONV = UnfortunatelyYes
>         NO_REGEX = NeedsStartEnd
> 
> Could we get that into rc1?

> I now have a different issue:
>
> make -C t/ all
> make[1]: Entering directory '/home/git/git/t'
> rm -f -r 'test-results'
> /usr/coreutils/bin/bash: /usr/bin/perl: Argument list too long
>
> Is there anyway to move to xargs? I am not sure why /usr/bin/perl is
> having issues with the build.

> On Mon, Jan 10, 2022 at 10:51:39PM -0500, rsbecker@nexbridge.com
> wrote:
> > I now have a different issue:
> >
> > make -C t/ all
> > make[1]: Entering directory '/home/git/git/t'
> > rm -f -r 'test-results'
> > /usr/coreutils/bin/bash: /usr/bin/perl: Argument list too long
> >
> > Is there anyway to move to xargs? I am not sure why /usr/bin/perl is
> > having issues with the build.
> 
> Tracing through t/Makefile, I am _pretty_ sure that this Perl invocation comes
> from the test-lint-shell-syntax recipe.
> 
> I wonder if something like this would do the trick?
> 
> --- 8< ---
> 
> diff --git a/t/Makefile b/t/Makefile
> index 46cd5fc527..d959119133 100644
> --- a/t/Makefile
> +++ b/t/Makefile
> @@ -90,7 +90,7 @@ test-lint-executable:
>  		echo >&2 "non-executable tests:" $$bad; exit 1; }
> 
>  test-lint-shell-syntax:
> -	@'$(PERL_PATH_SQ)' check-non-portable-shell.pl $(T) $(THELPERS)
> $(TPERF)
> +	@echo $(T) $(THELPERS) $(TPERF) | xargs '$(PERL_PATH_SQ)'
> +check-non-portable-shell.pl
> 
>  test-lint-filenames:
>  	@# We do *not* pass a glob to ls-files but use grep instead, to catch
> 
> --- >8 ---

> diff --git a/config.mak.uname b/config.mak.uname
> index a3a779327f..9b3e9bff5f 100644
> --- a/config.mak.uname
> +++ b/config.mak.uname
> @@ -576,6 +576,7 @@ ifeq ($(uname_S),NONSTOP_KERNEL)
>         NO_SETENV = YesPlease
>         NO_UNSETENV = YesPlease
>         NO_MKDTEMP = YesPlease
> +       NO_UNCOMPRESS2 = YesPlease
>         # Currently libiconv-1.9.1.
>         OLD_ICONV = UnfortunatelyYes
>         NO_REGEX = NeedsStartEnd
>
> Could we get that into rc1?

> <rsbecker@nexbridge.com> writes:
>
> > diff --git a/config.mak.uname b/config.mak.uname
> > index a3a779327f..9b3e9bff5f 100644
> > --- a/config.mak.uname
> > +++ b/config.mak.uname
> > @@ -576,6 +576,7 @@ ifeq ($(uname_S),NONSTOP_KERNEL)
> >         NO_SETENV = YesPlease
> >         NO_UNSETENV = YesPlease
> >         NO_MKDTEMP = YesPlease
> > +       NO_UNCOMPRESS2 = YesPlease
> >         # Currently libiconv-1.9.1.
> >         OLD_ICONV = UnfortunatelyYes
> >         NO_REGEX = NeedsStartEnd
> >
> > Could we get that into rc1?
>
> Sure, with an appliable patch that is properly signed off.

> <rsbecker@nexbridge.com> writes:
>
>> diff --git a/config.mak.uname b/config.mak.uname
>> index a3a779327f..9b3e9bff5f 100644
>> --- a/config.mak.uname
>> +++ b/config.mak.uname
>> @@ -576,6 +576,7 @@ ifeq ($(uname_S),NONSTOP_KERNEL)
>>         NO_SETENV = YesPlease
>>         NO_UNSETENV = YesPlease
>>         NO_MKDTEMP = YesPlease
>> +       NO_UNCOMPRESS2 = YesPlease
>>         # Currently libiconv-1.9.1.
>>         OLD_ICONV = UnfortunatelyYes
>>         NO_REGEX = NeedsStartEnd
>>
>> Could we get that into rc1?
>
> Sure, with an appliable patch that is properly signed off.
>
> ----- >8 --------- >8 --------- >8 --------- >8 --------- >8 -----
> Date: Mon, 10 Jan 2022 22:51:39 -0500
> From: "Randall S. Becker" <rsbecker@nexbridge.com>
> Subject: [PATCH] build: NonStop ships with an older zlib
>
> Notably, it lacks uncompress2(); use the fallback we ship in our
> tree instead.
>
> not-Signed-off-yet-by: Randall S. Becker <rsbecker@nexbridge.com>
> Signed-off-by: Junio C Hamano <gitster@pobox.com>
> ---
>  config.mak.uname | 1 +
>  1 file changed, 1 insertion(+)
>
> diff --git a/config.mak.uname b/config.mak.uname
> index a3a779327f..9b3e9bff5f 100644
> --- a/config.mak.uname
> +++ b/config.mak.uname
> @@ -576,6 +576,7 @@ ifeq ($(uname_S),NONSTOP_KERNEL)
>  	NO_SETENV = YesPlease
>  	NO_UNSETENV = YesPlease
>  	NO_MKDTEMP = YesPlease
> +	NO_UNCOMPRESS2 = YesPlease
>  	# Currently libiconv-1.9.1.
>  	OLD_ICONV = UnfortunatelyYes
>  	NO_REGEX = NeedsStartEnd