Re: [patch] add gdb.threads/pthread_cond_wait.* in fnchange.lst

Re: [patch] add gdb.threads/pthread_cond_wait.* in fnchange.lst

[Michael Elizabeth Chastain]

Hi Joel,

My understanding is that the test suite does not actually *run*
on djgpp.  So the files just need to un-tar cleanly, which means
only one '.' in the filename and no '+' and the default truncation
to 8.3 has to give unique names.  So your files would have unpacked
to testsuit/gdb.thr/pthread_.exp and testsuit/gdb.thr/pthread_.c
and that would be okay (just barely, because of pthreads.exp).

Eli is the expert on this ... Eli?

Michael C

Re: [patch] add gdb.threads/pthread_cond_wait.* in fnchange.lst

[Eli Zaretskii]

> Date: Wed, 19 May 2004 22:57:54 -0400 (EDT)
> From: mec.gnu@mindspring.com (Michael Elizabeth Chastain)
> 
> My understanding is that the test suite does not actually *run*
> on djgpp.  So the files just need to un-tar cleanly, which means
> only one '.' in the filename and no '+' and the default truncation
> to 8.3 has to give unique names.  So your files would have unpacked
> to testsuit/gdb.thr/pthread_.exp and testsuit/gdb.thr/pthread_.c
> and that would be okay (just barely, because of pthreads.exp).
> 
> Eli is the expert on this ... Eli?

You are absolutely right.

Note that, while the test suite cannot be run with the DJGPP build
(because `expect' needs async subprocesses that DJGPP doesn't
support), even if it could be run, the truncated names would work.
That is because, when a program tries to open a file whose name is
pthread_cond_wait.exp, the OS silently truncates that to the 8+3
limits, yielding pthread_.exp and looks for a file by that truncated
name.  That's why we can compile GDB itself (which involves the
compiler, Make, the linker, etc. having to open the source files),
although many of its source files have names in excess of 8+3 limits.

Re: [patch] add gdb.threads/pthread_cond_wait.* in fnchange.lst

[Joel Brobecker]

> > So if I understand it right, the additions I just made in
> > fnchange.lst where not needed. In that case, I think it would be
> > better if I removed them. What do you think?
> 
> Yes, please remove them.

Removed by the following patch:

2004-05-21  Joel Brobecker  <brobecker@gnat.com>

        * config/djgpp/fnchange.lst: Undo previous change, was useless.

-- 
Joel
Attachment:
condwait.diff
Description: Text document

Re: [patch] add gdb.threads/pthread_cond_wait.* in fnchange.lst

[Eli Zaretskii]

> Date: Thu, 20 May 2004 09:27:04 -0700
> From: Joel Brobecker <brobecker@gnat.com>
> 
> So if I understand it right, the additions I just made in
> fnchange.lst where not needed. In that case, I think it would be
> better if I removed them. What do you think?

Yes, please remove them.

Re: [patch] add gdb.threads/pthread_cond_wait.* in fnchange.lst

[Joel Brobecker]

> The OS truncates the names to 8+3 limits automatically when the
> tarball is untarred; that truncation is an integral part of the
> operation of the DOS system calls which work with file names.  So the
> only 2 problems that we need to fix with entries in fnchange.lst are:

Thanks for the very detailed explaination. So if I understand it right,
the additions I just made in fnchange.lst where not needed. In that
case, I think it would be better if I removed them. What do you think?

-- 
Joel

Re: [patch] add gdb.threads/pthread_cond_wait.* in fnchange.lst

[Eli Zaretskii]

> Date: Wed, 19 May 2004 15:00:40 -0700
> From: Joel Brobecker <brobecker@gnat.com>
> 
>         * config/djgpp/fnchange.lst: Add entries for pthread_cond_wait.exp
>         and pthread_cond_wait.c.
> 
> I must admit that i was confused a bit by what I saw. For instance,
> I saw that we didn't translate the "testsuite" part in the path,
> although it has 9 characters. Nor did we translate directory names
> such as "gdb.base" or "gdb.arch" (extension has more than 3 chars).

The OS truncates the names to 8+3 limits automatically when the
tarball is untarred; that truncation is an integral part of the
operation of the DOS system calls which work with file names.  So the
only 2 problems that we need to fix with entries in fnchange.lst are:

  - When 2 or more files produce the same name after truncation to
    8+3;

  - When a file's name includes characters that are disallowed in DOS
    file names.  Examples of such characters include `+', `,', `:',
    and a few more.  File names with leading dots, like .gdbinit, are
    also not allowed.

The fnchange.lst file is submitted to a program used to untar the
distribution (see djunpack.bat for the gory details), and which
performs the renaming on the fly.  That program automatically renames
some of the files whose names are invalid on DOS and are frequently
used in GNU software, like foo.c++, .foobar, etc.  This is another
reason that some invalid names might be missing from fnchange.lst.

[patch] add gdb.threads/pthread_cond_wait.* in fnchange.lst

[Joel Brobecker]

Hello,

I just made the following change:

2004-05-19  Joel Brobecker  <brobecker@gnat.com>

        * config/djgpp/fnchange.lst: Add entries for pthread_cond_wait.exp
        and pthread_cond_wait.c.

I must admit that i was confused a bit by what I saw. For instance,
I saw that we didn't translate the "testsuite" part in the path,
although it has 9 characters. Nor did we translate directory names
such as "gdb.base" or "gdb.arch" (extension has more than 3 chars).
So my understanding was that we only needed to translate the filename
to 8+3...

-- 
Joel
Attachment:
fnchange.lst.diff
Description: Text document