Re: Solaris and sed: Too many commands, last: s/\n//

From: Jeffrey Walton <noloader@gmail.com>
Date: Mar 12, 2019, 03:57 UTC
Message-ID: <CAH8yC8=gsuknMOAYyjVDyZENDiJoRia+MnwhsMhZdcnfuO=cbw@mail.gmail.com>
In-Reply-To: <20190312024542.GA7411@flurp.local>

On Mon, Mar 11, 2019 at 10:45 PM Eric Sunshine <sunshine@sunshineco.com> wrote:

Show 49 quoted lines

>
> On Mon, Mar 11, 2019 at 10:11:41PM -0400, Jeffrey Walton wrote:
> > On Mon, Mar 11, 2019 at 9:55 PM Jeffrey Walton <noloader@gmail.com> wrote:
> > > Oh man, you're using GNU make. I thought Git was using that anemic
> > > Posix Make. See attached.
> > >
> > > I think Solaris provides an older gawk. Is this an easier problem:
> > >
> > >     awk: chainlint.sed:88: :squash
> > >     awk: chainlint.sed:88: ^ syntax error
> > >     awk: chainlint.sed:91:  s/\\\n//
> > >     awk: chainlint.sed:91:    ^ backslash not last character on line
> > >     Usage: awk [POSIX or GNU style options] -f progfile [--] file ...
> > >     Usage: awk [POSIX or GNU style options] [--] 'program' file ...
> >
> > My bad , there was a typo... 'awk' got assigned to SED variable.
> >
> > This patch works as expected.
> >
> > diff --git a/t/Makefile b/t/Makefile
> > @@ -11,11 +11,25 @@ SHELL_PATH ?= $(SHELL)
> >  TAR ?= $(TAR)
> > +AWK ?= $(AWK)
> > +SED ?= $(SED)
> > +GREP ?= $(GREP)
> >
> > +# Fix Solaris tools. These are Posix. GNU tools located at /usr/gnu/bin.
> > +ifneq ($(wildcard /usr/gnu/bin/grep),)
> > +  GREP := /usr/gnu/bin/grep
> > +endif
> > +ifneq ($(wildcard /usr/gnu/bin/sed),)
> > +  SED := /usr/gnu/bin/sed
> > +endif
> > +ifneq ($(wildcard /usr/gnu/bin/awk),)
> > +  SED := /usr/gnu/bin/awk
> > +endif
>
> I think the last assignment ought to be "AWK := ...".
>
> Anyhow, this sort of platform-specific tool customization is typically done by config.mak.uname in the top-level Git directory. In fact, there's already a section for SunOS:
>
>     ifeq ($(uname_S),SunOS)
>         ...
>         SANE_TOOL_PATH = /usr/xpg6/bin:/usr/xpg4/bin
>         ...
>
> Prepending /usr/gnu/bin to SANE_TOOL_PATH might be a good idea as a first step toward fixing the problem you're seeing on Solaris, however, as Ćvar mentioned in [1], SANE_TOOL_PATH isn't presently consulted when running tests. But, as he suggested, fixing the tests to respect SANE_TOOL_PATH might be a good solution overall.
>
> So, rather than making platform-specific customizations to t/Makefile, an arguably better solution would be to update config.mak.uname to add /usr/gnu/bin to SANE_TOOL_PATH and then update the test system to respect that value (thus, these GREP, SED, AWK specializations can be avoided).

I may have been blowing away SANE_BUILD_PATH.

My build script calls:

   SANE_BUILD_PATH="<my updated tool path> \
   PKG_CONFIG_PATH="..." \
   ... \
./configure \
    --prefix=<my new tool location> \
    --libdir="..." \
    ...

I set SANE_BUILD_PATH in the environment because I want Git to use the new tools being built for testing.

Or, if Git is blowing away my SANE_BUILD_PATH , then that might explain the encoding failures. The updated tools, like new iConv and Unistring, are not on PATH. The updated tools are located at /var/sanitize/bin, and are passed through SANE_BUILD_PATH .

This might also explain the intermittent crashes I see. Old tools from /usr/bin are used, but new libraries from /var/sanitize/lib are used because of LD_LIBRARY_PATH. Gotta love those fucking path problems that have plagued Linux for the last 30 years. The idiot who figured it was a good idea to compile and link against one library, and then load the wrong library at runtime, should get a platinum Darwin award.

It may be better to set SANE_TOOL_PATH something like:

    SANE_TOOL_PATH = "$SANE_TOOL_PATH:/usr/xpg6/bin:/usr/xpg4/bin"

Jeff