Hi everybody,

I have uploaded the last alpha binary for djgpp gpc-980401.i386-pc-msdos-djgpp.zip and found some bugs.

First I have updated gcc/gpp/lgpp to version 2.80 following instructions from djgpp. Since I program nearly always in pascal, it was pointless to update before. First clean removal of gpc-971001, and versions 2.7.2.1 of gcc/gpp/lgpp using data contained in the manifest directory, then installation of the binaries for 2.80. Modification of the djgpp.env file to reflect the new order of directory search: in particular removal of the [gcc] and [cpp] sections. Modification of the Specs file in the lib directory to include -remap in the *cpp: section. Then I ran some test C programs of mine, and recompile grx22 and bccgrx20 libraries with their test programs, and run them. All is OK. Then I installed the gpc djgpp binaries. I had to make decisions about the djgpp.env (removing the [gpc] and [gpc-cpp] sections analogous to the [gcc] and [cpp] sections) and specs file (the specs.gpc included in 971001 was no more in 980401), but this has no influence on the bug. The most conspicuous bug occurs each time a program calling a unit is compiled with automake. It is displayed in the attached couple of short program, taken for your old test directory, with some adaptation, because the bug has nothing to do with what was the purpose of the test. (Other programs from the same test directory run OK). All is made under plain DOS/ DJGPP (no windows above head): Compiled with:

redir -e bug.lst gpc -g --automake=-g -o bug.exe bug.pas

which gives the following bug.lst file:

bug.pas:15: warning: missing program header bug.pas:15: only one program declaration allowed bug.pas:15: syntax error before `Ok'

Notice that a second make gives no more error and that the resulting program is then OK. To give you more information I have attached also the files bug1.lst produced with the -v option for gpc for the first compilation, and bug2.lst produced in the same conditions for the second compilation (which works).

In a dos box under windows 3.11, there is also a bug in the same conditions, but the message in bug.lst is completely different:

gpc1.exe: C:\TMP/ccaaaaaa.s: I/O error

similar bug files bugw1.lst and bugw2.lst are also attached.

Other bugs are there, but some may be consequence of the previous, which appear eveywhere, (and all my programs use units).

On Wed, 1 Apr 1998, Peter Gerwinski wrote:

Hi Peter,

Hello everybody,

a new alpha (gpc-980401) is being uploaded to ftp://agnes.dida.physik.uni-essen.de/gnu-pascal/alpha/. This time I also have binaries for Linux and DJGPP (in the subdirectory `binary').

the new alpha compiles under IRIX 6.2 with gcc-2.8.0. However the resulting binary seems to be crap.

0) - Copy the gpc source under the gcc tree.

1) - this version of gcc already support IRIX 6.2 while gas still won't. Thus the configure script should started without any additional switch:

% ./configure % cd p % ./configure % cd ..

2) - Build gcc:

make bootstrap LANGUAGES=c BOOT_CFLAGS=-O

3) - Build gpc:

make CC="stage2/xgcc -Bstage2/" CFLAGS=-O LANGUAGES=pascal

It compiles without much trouble unfortunately the binary fails on the tests:

% cd p/test/extended_pascal % make PC="../../../xgpc -B../../../"

You get:

../../../xgpc -B../../../ -g -o abs abs.pas xgpc: Internal compiler error: program gpc1 got fatal signal 6 make: *** [abs] Error 1

That's it. Still a bit to do.....

Cheers,

miklos

Miklos Cserzo University of Birmingham c/o School of Biochemistry MRC - Bioinformatics Project Tel: +44-121-414-4090 Schools of Biochemistry, Biology and Medicine Fax: +44-121-414-3982 Edgbaston, Birmingham B15 2TT E-mail: miklos@bip.bham.ac.uk United Kingdom