[tlbuild] texlive 2011 build on aix 4.3 - "make check": 12 of 46 tests failed

[Vladimir Volovich]

"VV" == Vladimir Volovich writes:

 VV> I've bisected the source tree between those revisions (compiling
 VV> and testing if it works, narrowing the range of revisions each
 VV> time), and found an exact time when it happened:

 VV> last revision where TeX still successfully runs the trip test, is:

 VV> r12461 | peter | 2009-03-20 13:24:22 +0300 (Fri, 20 Mar 2009) | 1 line

 VV> and it starts to segfault in

 VV> r12464 | peter | 2009-03-20 13:47:15 +0300 (Fri, 20 Mar 2009) | 1 line

 VV> Further, i've compared the generated C sources of TeX (texini.c,
 VV> tex0.c, tex1.c, tex2.c, tex-pool.c, texextra.c), between the good
 VV> revision r12461 and the bad revision r12464, and I see the
 VV> following changes:
[...]
 VV> i.e. -10000 was changed to 2550000L in a few places, and this broke
 VV> TeX on AIX.

moreover, compiline sources from r12464 results in buggy TeX executable
(which segfaults while processing trip.tex), but if I just revert the
changes to texini.c and tex1.c which I've posted in the previous email,
it gives a working TeX.

so the breakage was definitely caused by this change from -10000 to 2550000L
in the web2c-generated sources.

now we need to understand:

1) why was this change made? was it intentional? does it occur on all
   platforms or only on AIX because of some change to web2c which
   manifests only on AIX (i.e. do texini.c and tex1.c generated on other
   platforms use -10000 or 2550000L in the same places?).

2) why/how does it result in a segfault on AIX, but not on other platforms.

Best wishes,
v.