Maurice Lombardi wrote:
Mmm. There is something broken in gpc under gcc-2.9.5.3. It compiles with snapshot gpc-20010315, but it gives lots of errors when running the test suite. As a comparison I have compiled the same snapshot under gcc-2.9.5.2 (the result is on agnes) and it gives zero error for the test suite.
Briefly speaking:
when applying the patch contained in the p/diff directory (I have taken the same diff as for gcc-295 -2951 -2952 which are identical), I get the following messages
C:\djgpp\gnu\gcc-2.953\gcc>patch -p1 < p\diffs\gcc-2.95.3.diff patching file "expr.c" Hunk #1 succeeded at 4505 (offset 75 lines). Hunk #3 succeeded at 4542 (offset 75 lines). patching file "fold-const.c" Hunk #1 succeeded at 1462 (offset 1 line). patching file "stor-layout.c" patching file "tree.c" Hunk #1 succeeded at 5025 (offset 39 lines). Hunk #3 succeeded at 5100 (offset 39 lines). patching file "tree.h" Hunk #1 succeeded at 1631 (offset 1 line). patching file "tree.def"
No hunk fails, so I proceed with a 3 stage bootstrap. Seems OK but when running dostest it crashes midway, after lots of errors the log file is on agnes:
ftp://agnes.dida.physik.uni-essen.de/home/maurice/make.2953.out
It appears that:
all failed programs include units, either by an explicit uses clause in the main, by a --uses= compilator option or even throug $L there are error messages gpc.exe: installation problem, cannot exec `cpp': No such file or directory (ENOENT)
That happens if the programs use C files (directly or indirectly, e.g. via CRT), or does it also happen if they just use Pascal units?
indeed cpp has been renames cpp0 in this release.
I haven't looked at gcc-2.95.3 yet, but you might want to try replacing `cpp' by `cpp0' in gpc.c, lines 713, 749, 777, 795, 826, 862, 884, 897, 921, 949.
The other problems (to me) don't look cpp related. Maybe the diffs went wrong, or there are more serious differences. Probably only Peter can tell...
Frank