Re: [Gcl-devel] 2.6.2.....

[Mike Thomas]

Hi all.

For the purposes of GCL 2.6.2 I think the Maxima "ignore-errors" bug is 
nailed.

Removing all but the directory separator character conversion in 
"fix_filename" in "o/mingwin.c" removes the problem.  The removed 
portions are redundant cygwin path conversion which I should have 
flicked ages ago and device working directory code which seems to have a 
number of potentially catastrophic holes in it, and which is arguably 
not necessary anyway.  Both of these removed parts copy strings around.

This function is called in a number of places, including the Lisp 
functions "namestring" and "truename" which are well used in Maxima's 
"defsystem.lisp".

Even though "fix_filename" does not allocate it's own memory, it does 
mess with Lisp string object storage and is not wrapped in 
"BEGIN/END_NO_INTERRUPT" which is probably how the randomisation occurs.

Vadim and Dave, could I please ask you if at all possible to replace the 
function "fix_filename" in "o/mingwin.c" with this one:

void
fix_filename( object pathname, char *filename )
{
    char *p = filename;
    while ( *p ) {
        if (*p=='\\') *p='/';
        p++;
    }
}

and let us know whether the ignore-errors bug goes away under conditions 
which would normally cause it for you?

For me those conditions are the use of top level gcc optimisation of 
less than "-O3" in "configure.in" (don't forget to "autoconf" after 
altering "configure.in") or just to configure a GCL build with 
"--enable-debug".

I've tested both former routes to disaster and found them fixed. 
Hopefully I will have similar success on my other build machine as will 
each of you on your's.

I can already feel the weight beginning to lift.  Millions of Windows 
users the world over can rest easier tonight.

Good night

Mike Thomas.