Possible GNU Make bug(?)

[Tim Noell]

Hi GNU Make Team:

I have encountered a situation with GNU Make that doesn't seem right.  It could 
be something simple that I am overlooking, but wanted to forward the example so 
you folks could check it out.

Attached please find gnumakebug.tar.gz, containing my minimum example.

I am running GNU Make 3.79.1 on Solaris 5.8 on a Sun Ultra-5_10.  
Unfortunately, 
I do not have the config.h file from when the administrators built the 
executable.

The Makeifle has a simple target "echoit" that echoes a string that is built up 
by two included makefiles (inclfile[12].mk) that are made from their "sources" 
(inclfilep[12].stub).  At the time of invocation of make, the files 
inclfile[12].mk do not exist.  The twist here is that the rule that is used to 
make the *.mk files ( the %.mk :   rule), makes BOTH *.mk files.

However, subsequently, make decides it needs to make the second one again.

The file make.out is the output of the commnad
make -d echoit >& make.out

In make.out, lines 26-31, both files are made due to make making inclfile2.mk.

Subsequently, on line 37, make says that inclfile1.mk does not exist!

I thought this might be a parallelism problem, but the rule for creating the 
*.mk files is a single line, so it should be executed in a single shell.
For grins I tried adding a .NOTPARALLEL target, but this had no effect.

I would really like to get a fix and/or work around for this problem, as it is 
causing me a non-scalable performance hit with a growing re-usable component 
build environment.

Thanks for any and all help,
Tim

// Tim Noell  
// Specialist, Business Printer Division
// Lexmark International, Inc.  
// "Fish heads, fish heads, rolly polly fish heads ..."

gnumakebug.tar.gz
Description: gnumakebug.tar.gz