Old bug: the 14-character limit.

Old bug: the 14-character limit.

[Nathanael Nerode]

I was going through old bugs and I found one (#4284) about filenames 
which aren't unique within 14 characters.

Are we still trying to preserve 14-character/8.3 uniqueness for building 
GCC (at least for C; C++ and Java libraries have other standards)?  I 
was under the impression that we were, but I couldn't find it 
documented anywhere.  If we are, can we document it?  If we aren't, can 
we document it and close the bug?

--Nathanael

Re: Old bug: the 14-character limit.

[DJ Delorie]

> Are we still trying to preserve 14-character/8.3 uniqueness for
> building GCC (at least for C; C++ and Java libraries have other
> standards)?  I was under the impression that we were, but I couldn't
> find it documented anywhere.

I don't think there are any old SYSV systems still around that have
the 14 character limit.

For DJGPP (the one remaining 8.3 system), we've accepted the fact that
an LFN-capable system is required for building, but still require 8.3
compatibility in the *installed* files (i.e. you need LFN to build,
but not to use).  This includes C++ and Java.

Note that all Microsoft-based systems also have case-insensitive file
systems, so things like String.h and string.h are the same file.

The FSF has a program called "doschk" that can scan for potential file
name conflicts.

Re: Old bug: the 14-character limit.

[Andris Pavenis]

On Friday 23 May 2003 22:51, DJ Delorie wrote:
> > Are we still trying to preserve 14-character/8.3 uniqueness for
> > building GCC (at least for C; C++ and Java libraries have other
> > standards)?  I was under the impression that we were, but I couldn't
> > find it documented anywhere.
>
> I don't think there are any old SYSV systems still around that have
> the 14 character limit.
>
> For DJGPP (the one remaining 8.3 system), we've accepted the fact that
> an LFN-capable system is required for building, but still require 8.3
> compatibility in the *installed* files (i.e. you need LFN to build,
> but not to use).  This includes C++ and Java.

For C++ includes a part of files which names are incompatible with 
8+3 system (eg. ones which contains '+', or may not be unique on such system)
are renamed and corresponding header.gcc files is generated to provide 
possibility to use the original names. Also the name of directory where C++ 
includes are installed is changed.

Currently Java is not supported for DJGPP (perhaps compiler would still
build, but runtime is not ported at all). I don't see possibility of changes 
unless somebody will work on this.

Andris

Re: Old bug: the 14-character limit.

[Laurent GUERBY]

According to my experiments, the longest file name in the gcc source
tree is 48 characters: gnu_java_awt_peer_gtk_GtkCheckboxMenuItemPeer.c
in libjava/jni/gtk-peer/.

Is there a documented upper limit on file name length for the gcc tree?
If not is there a common wisdom upper limit?

Laurent

Re: Old bug: the 14-character limit.

[Neil Booth]

Laurent GUERBY wrote:-

> According to my experiments, the longest file name in the gcc source
> tree is 48 characters: gnu_java_awt_peer_gtk_GtkCheckboxMenuItemPeer.c
> in libjava/jni/gtk-peer/.
> 
> Is there a documented upper limit on file name length for the gcc tree?
> If not is there a common wisdom upper limit?

There is now: 48 8-)  I think that's enough for us.

Neil.

Re: Old bug: the 14-character limit.

[Phil Edwards]

On Fri, May 23, 2003 at 10:22:01PM +0200, Laurent GUERBY wrote:
> According to my experiments, the longest file name in the gcc source
> tree is 48 characters: gnu_java_awt_peer_gtk_GtkCheckboxMenuItemPeer.c
> in libjava/jni/gtk-peer/.

Like DJ said, we don't care about name length for building, only for
installing, and most of us don't even care very much about that.  :-)

> Is there a documented upper limit on file name length for the gcc tree?
> If not is there a common wisdom upper limit?

Some versions of (non-GNU) tar have a 100-char limit on certain names.

-- 
If ye love wealth greater than liberty, the tranquility of servitude greater
than the animating contest for freedom, go home and leave us in peace.  We seek
not your counsel, nor your arms.  Crouch down and lick the hand that feeds you;
and may posterity forget that ye were our countrymen.            - Samuel Adams

Re: Old bug: the 14-character limit.

[Mike Stump]

On Friday, May 23, 2003, at 01:22 PM, Laurent GUERBY wrote:
> If not is there a common wisdom upper limit?

250 works just fine...  260 won't work on a lot of systems.  :-)

Re: Old bug: the 14-character limit.

[DJ Delorie]

> Is there a documented upper limit on file name length for the gcc tree?

I don't know of any hard limits, but as long as files are unique in
the 8.3 characters (i.e. first 8 characters of the basename, first 3
characters of the extension), even DJGPP won't care about the length.

IIRC Windows has a hard upper limit on total path size, but I forget
what it is.  DJGPP's MAXPATH is 260 (for LFN; 80 for SFN), which is
probably related to the Windows limit.

Note that some archive formats (ar) have object name length limits.

Re: Old bug: the 14-character limit.

[Laurent GUERBY]

On Fri, 2003-05-23 at 22:53, DJ Delorie wrote:
> I don't know of any hard limits, but as long as files are unique in
> the 8.3 characters (i.e. first 8 characters of the basename, first 3
> characters of the extension), even DJGPP won't care about the length.

In the case I mentionned, there are multiple files with the same 25
first characters and the same extension, so I guess it's out of DJGPP
league anyway :).

I assume DJGPP constaints are mostly for the C compiler (may be other
languages ?), is this documented somewhere?

Laurent

Re: Old bug: the 14-character limit.

[DJ Delorie]

> In the case I mentionned, there are multiple files with the same 25
> first characters and the same extension, so I guess it's out of DJGPP
> league anyway :).

Only if it needs to be installed.

And I fail to see the humor in excluding DJGPP as a platform.  GCC has
supported DJGPP (or at least been built for it) since version 1.35.

> I assume DJGPP constaints are mostly for the C compiler (may be other
> languages ?), is this documented somewhere?

We support all the languages except Java at the moment[*].  I suspect
the reason for not supporting Java is the runtime bits, though.

[*] I see c, c++, ada, pascal, objc, and fortran on the ftp site.
    That's for 3.2.3; 3.3 is the same except no pascal yet.

Re: Old bug: the 14-character limit.

[Laurent GUERBY]

On Fri, 2003-05-23 at 23:31, DJ Delorie wrote:
> > In the case I mentionned, there are multiple files with the same 25
> > first characters and the same extension, so I guess it's out of DJGPP
> > league anyway :).
> 
> Only if it needs to be installed.
> 
> And I fail to see the humor in excluding DJGPP as a platform.  GCC has
> supported DJGPP (or at least been built for it) since version 1.35.

I'm just trying to understand what are the requirements to follow,
I asked if they were documented, since I had no reply I assume they
aren't (undocumented requirements are obviously harder to follow :). If
I do some guesswork on the replies so far:

- As far as we're not too far from 100 (portential tar limit) or 250
characters (more common limit on modern file systems)
for files needed during the build, and that includes DJGPP, right?
(I don't know DJGPP, so I assume there's some distinction
between build and use once installed.)

- If one wants to install a compiler on a platform, 8+3 name uniqueness
for all files to be installed is required for the compiler to be
useable under the DJGPP platform.

Laurent

Re: Old bug: the 14-character limit.

[DJ Delorie]

> - As far as we're not too far from 100 (portential tar limit) or 250
> characters (more common limit on modern file systems)
> for files needed during the build, and that includes DJGPP, right?
> (I don't know DJGPP, so I assume there's some distinction
> between build and use once installed.)

Yes.  The thing about DJGPP is it supports two filesystem APIs, "short
file names" (SFN, or 8.3) and "long file names" (LFN).  Not all
Windows OSs provide LFN services, and some DOS systems have LFN
drivers.  In general, though, it's acceptable to require that someone
BUILDING a gnu package make the effort to obtain LFN services.

> - If one wants to install a compiler on a platform, 8+3 name
> uniqueness for all files to be installed is required for the
> compiler to be useable under the DJGPP platform.

Yes, because for USING a gnu package it is unacceptable to require LFN
services.  Dosemu, for example, cannot provide them for a shared linux
partition.

Re: Old bug: the 14-character limit.

[Michael S. Zick]

On Friday 23 May 2003 03:53 pm, DJ Delorie wrote:
> > Is there a documented upper limit on file name length for the gcc tree?
>
> I don't know of any hard limits, but as long as files are unique in
> the 8.3 characters (i.e. first 8 characters of the basename, first 3
> characters of the extension), even DJGPP won't care about the length.
>
> IIRC Windows has a hard upper limit on total path size, but I forget
> what it is.  DJGPP's MAXPATH is 260 (for LFN; 80 for SFN), which is
> probably related to the Windows limit.
Some GNU/Linux systems use:

[root@wolf686 mszick]# stat -f /dev/hda
  File: "/dev/hda"
    ID: 0        0        Namelen: 255     Type: EXT2

Re: Old bug: the 14-character limit.

[Nathanael Nerode]

That should read bug 4293. *cough*.