Bug 294543 – sci-electronics/iverilog failt to compile when usign parallel compile (-j2)

• Product: Gentoo Linux
• Status: RESOLVED FIXED

[Description Piotr Stachura 2009-11-25 08:25:00 UTC]

Compile fails when parallel compiling.
Compiling of main.cc includes not yet generated version.h file:

cat ./version_base.in version_tag.h > version.h
x86_64-pc-linux-gnu-g++ -DHAVE_CONFIG_H -I. -I.  -Wall -march=k8 -Os -pipe -m3dnow -MD -c main.cc -o main.o
x86_64-pc-linux-gnu-g++ -DHAVE_CONFIG_H -I. -I.  -Wall -march=k8 -Os -pipe -m3dnow -MD -c async.cc -o async.o
mv async.d dep/async.d
x86_64-pc-linux-gnu-g++ -DHAVE_CONFIG_H -I. -I.  -Wall -march=k8 -Os -pipe -m3dnow -MD -c design_dump.cc -o design_dump.o
main.cc: In function ‘int main(int, char**)’:
main.cc:611: error: ‘VERSION’ was not declared in this scope
main.cc:611: error: ‘VERSION_TAG’ was not declared in this scope

Reproducible: Always

Steps to Reproduce:
1. Set MAKEOPTS="-j2" in make.conf
2. Run compile of iverilog-0.9.1
3. Watch fail

[Comment 1 Piotr Stachura 2009-11-25 08:26:05 UTC]

Created attachment 211125 [details, diff]
Proposed patch

Force make to use "-j1"

[Comment 2 Thomas Beierlein 2010-03-04 17:05:08 UTC]

While I can not reproduce the mentioned problem on some different machines with -j2..-j5 there really is a problem in the Makefile.in.

Line 74 from Makefiel.in reads
all: dep config.h _pli_types.h version.h ivl

In a parallel make it starts the creation of version.h and the compilation of all in parallel. Some of the sources needs version.h (see reported error).

The following patch request for all sources (hopefully), that version.h gets build 'before' any compilation of *.cc files to *.o.

That fixes the problem for me, but maybe there are some more problems....

[Comment 3 Thomas Beierlein 2010-03-04 17:07:58 UTC]

Created attachment 222045 [details, diff]
Fixe for iverilog-0.9.1.ebuild to fix parallel make issue in Makefile.in

[Comment 4 Michael Weber (RETIRED) 2012-05-25 13:31:30 UTC]

It was fixed by an -j in versions 0.9.1 and 0.9.3.

I've added 0.9.5 today without emake -j1, it doesn't exhibit this problem - so far.