bug report

[Francis Grolemund]

Duh.  I almost forgot.  We're using version 3.78.1 of GNUMake.



Francis Grolemund
Edge Server Development
IBM Pittsburgh Lab
(412) 667-4989
T/L:  989-4989
address@hidden
---------------------- Forwarded by Francis Grolemund/Pittsburgh/IBM on
07/16/2001 09:52 PM ---------------------------

Francis Grolemund
07/16/2001 09:47 PM

To:   address@hidden
cc:
From: Francis Grolemund/Pittsburgh/address@hidden
Subject:  bug report


I'm sorry this isn't in the proper bug reporting format, but time is pretty
critical right now on our project.

In any case, we ran into a subtle bug that was partially caused by our
environment and how we've used gnumake, but I figured that since its such a
pain in the ass, you guys may want to know about it.  The problem is only
occurring on a Windows 2000 system.  To give some background, we've written
a sort of macro language that created GNUMakefiles automatically to make it
easier for developers to write the makefiles.  In some cases developers
were exporting environment variables that were devoid of a value.  So in
other words, when arr2envblk() in misc.c started to accumulate them into a
block of memory for process_begin() in sub_proc.c, some would be added in
the form:
          foo=<null>
          bar=<null>
          baz=<null>

Apparently CreateProcess really dislikes this because it would give a most
unpleasant error, returning an ERROR_MORE_DATA from GetLastError.  Although
this makes sense now, the fact that it is undocumented caused weeks of pain
for us.

My solution was to check for variables with no value and not include them
in the environment block and the problem went away.

I hope this helps.  Keep up the good work.

Fran




Francis Grolemund
Edge Server Development
IBM Pittsburgh Lab
(412) 667-4989
T/L:  989-4989
address@hidden