Re: [bug-rcs] rcs-5.7.96 build feedback

[Thien-Thi Nguyen]

() "Nelson H. F. Beebe" <address@hidden>
() Sat, 22 Jan 2011 16:18:17 -0700 (MST)

   I have a test lab of about 25 flavors of Unix on all of the major CPU
   architectures.  Builds and validations of rcs-5.7.96 have been
   successful on GNU/Linux, Mac OS X, FreeBSD, and NetBSD systems.

Cool, thanks for evaluating 5.7.96.  Recently, 5.7.97 became available,
so i would be very interested in finding out how that fares in your lab.

   However, there are build or test failures on several others.  Here is
   a problem summary:

OK, ready!

   ------------------------------------------------------------------------
   SGI IRIX MIPS

   No C99 compiler available.

I suppose we could try to relax the C99 compiler requirement (falling
back to C89) but this platform is such a low priority, i don't see it
realistically happening for RCS 5.8.  Nonetheless, the issue is
recorded: <https://savannah.gnu.org/bugs/?32215>, should someone wish to
invest some effort in the future.

   ------------------------------------------------------------------------

   Solaris 10 AMD64:

   gcc -std=gnu99 -std=gnu99 -DHAVE_CONFIG_H -I. -I../lib -I'../lib' 
-D_REENTRANT
   -g -O2 -MT b-feph.o -MD -MP -MF .deps/b-feph.Tpo -c -o b-feph.o b-feph.c
   b-feph.c: In function `init_ephemstuff':
   b-feph.c:171: error: prior parameter's size depends on 'count'

This error message mystifies me.  I see on line 171 the static func ‘reap’,
which has an arglist that includes ‘count’:

170  static void
171  reap (size_t count, struct sff all[count],
172        int (*cut) (char const *filename))

but the error message (1) names the wrong function; and (2) says "prior
parameter" when ‘sff’ is declared after (left to right).  Could this be
a bug in the compiler?

   ------------------------------------------------------------------------

   Solaris 10 AMD64 or SPARC:
   Compilation with "gcc-4.2.4 -std=gnu99" succeeds, but all tests fail:

   ./t010: syntax error at line 21: `(' unexpected
   FAIL: t010
   ./t030: syntax error at line 21: `(' unexpected
   FAIL: t030
   ...

   Changing SHELL to /bin/ksh, /bin/bash, /usr/local/bin/bash, etc does not 
help.

I suspect this is because that shell's ‘case’ statement does not support
a pattern starting with open paren, like:

case x in (x) echo ok ;; esac   # should be ok, but i see not everywhere

In the test suite such phrasing has now been converted to omit the open
paren: <http://git.savannah.gnu.org/cgit/rcs.git/commit/?id=f5bb5f4d99>.
For the record, this was one of those warnings i read but did not heed
(info "(autoconf) Limitations of Builtins") -- let my blunderings here
serve as an example!

   ------------------------------------------------------------------------

   Solaris 10 AMD64 SPARC:
   /usr/local/bin/gcc-4.3 -std=c99 -DHAVE_CONFIG_H -I. -I../lib -I'../lib'
   -D_REENTRANT -g -O2 -MT merger.o -MD -MP -MF .deps/merger.Tpo -c -o merger.o
   merger.c
   merger.c: In function 'merge':
   merger.c:98: error: 'label' undeclared (first use in this function)
   merger.c:98: error: (Each undeclared identifier is reported only once
   merger.c:98: error: for each function it appears in.)

Brain fade, fixed in repo:
<http://git.savannah.gnu.org/cgit/rcs.git/commit/?id=4b654e581b>.

I wonder what is a good way to incorporate testing for different
configure results (such as ‘!DIFF3_BIN’ in this case).

   ------------------------------------------------------------------------

   MirBSD IA-32:
   /usr/bin/mgcc -std=gnu99 -std=gnu99 -DHAVE_CONFIG_H -I. -I../lib -I'../lib'
   -I/usr/local/include -MT b-isr.o -MD -MP -MF .deps/b-isr.Tpo -c -o b-isr.o
   b-isr.c
   b-isr.c: In function `scratch_from_context':
   b-isr.c:119: error: structure has no member named `uc_stack'

   ------------------------------------------------------------------------
   OpenBSD IA-32:
   /usr/bin/cc -std=gnu99 -std=gnu99 -DHAVE_CONFIG_H -I. -I../lib -I'../lib'
   -I/usr/local/include -MT b-isr.o -MD -MP -MF .deps/b-isr.Tpo -c -o b-isr.o
   b-isr.c
   b-isr.c: In function `scratch_from_context':
   b-isr.c:119: error: structure has no member named `uc_stack'

I don't know what to do about these -- my understanding is that ‘struct
ucontext_t’ should have member ‘uc_stack’.  Could you please send me
manpages for sigaction(2), setcontext(2) and getcontext(2) for these
systems?

   ------------------------------------------------------------------------
   OpenSolaris 11 AMD64 with /bin/c99:

   Build succeeds, but 11 tests fail:
   FAIL: t160
   FAIL: t510
   FAIL: t511
   FAIL: t600
   FAIL: t601
   FAIL: t603
   FAIL: t604
   FAIL: t605
   FAIL: t606
   FAIL: t900
   FAIL: t999

   ------------------------------------------------------------------------

   OpenSolaris 11 AMD64 with /bin/cc:

   Build succeeds, but 8 tests fail:
   FAIL: t600
   FAIL: t601
   FAIL: t603
   FAIL: t604
   FAIL: t605
   FAIL: t606
   FAIL: t900
   FAIL: t999

   On that system, a build with /usr/sfw/bin/gcc (3.4.3) passes all tests.

Could you please send me a tarball of the build tree, if it's not too
heavy (for my ISP)?  If it's too heavy, i would appreciate seeing
specifically:

  config.log
  config.status
  tests/*.d/*    # that is, all files in the t???.d/ directories
                 # preserved for improperly behaving tests

plus a log of "make all check".


I look forward to any info you can provide, including a similar
evaluation of 5.7.97, towards a cleaner 5.8 release.  Thanks again.