Skip to contentSkip to conversation

Hard-coded gcc header path

DISCUSSION Jörn Engel 5 participants 9 messages Nov 11, 2008 Sparse checker

lore
Share & tools
lore
#0 lore raw

Sparse doesn't work for me when compiling userspace code. Others have experienced the same, so I refer to someone else's description of the symptom:
http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=505177

On my system, sparse tries several headers in order, neither of which
exists:
open("/usr/include/stddef.h", O_RDONLY) = -1 ENOENT (No such file or directory) open("/usr/local/include/stddef.h", O_RDONLY) = -1 ENOENT (No such file or directory) open("/usr/lib/gcc/i486-linux-gnu/4.1.2/include/stddef.h", O_RDONLY) = -1 ENOENT (No such file or directory)

Not a big surprise when looking at the headers that do exist:

Galway:/usr/lib/gcc/i486-linux-gnu# ll
total 20
drwxr-xr-x 4 root root 4096 2008-07-01 08:42 3.4.6
drwxr-xr-x 4 root root 4096 2007-09-23 17:59 4.0.4
drwxr-xr-x 3 root root 4096 2008-06-26 00:20 4.1
lrwxrwxrwx 1 root root    3 2007-09-23 20:57 4.1.3 -> 4.1
drwxr-xr-x 3 root root 4096 2008-07-13 11:06 4.2
lrwxrwxrwx 1 root root    3 2008-07-01 08:42 4.2.4 -> 4.2
drwxr-xr-x 4 root root 4096 2008-09-30 16:27 4.3
lrwxrwxrwx 1 root root    3 2008-07-01 08:42 4.3.1 -> 4.3
lrwxrwxrwx 1 root root    3 2008-08-08 18:18 4.3.2 -> 4.3

So why does sparse try 4.1.2 and not 4.1.3 or 4.3.2 or any other variant?

joern@Galway:/usr/src/kernel/sparse$ cat pre-process.h
#define GCC_INTERNAL_INCLUDE "/usr/lib/gcc/i486-linux-gnu/4.1.2/include"

Ah, it is hard-coding the path.

So what is the right solution to this problem? Call "gcc --version" and parse the (deliberately hard) output? Or make a copy of the gcc headers and ship them as /usr/lib/sparse/0.4.1/...?

Jörn

Signature · 7 lines
But this is not to say that the main benefit of Linux and other GPL
software is lower-cost. Control is the main benefit--cost is secondary.
-- Bruce Perens
--
To unsubscribe from this list: send the line "unsubscribe linux-sparse" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#1 lore raw
Quoted 6 lines · Jörn Engel

So why does sparse try 4.1.2 and not 4.1.3 or 4.3.2 or any other variant?

joern@Galway:/usr/src/kernel/sparse$ cat pre-process.h
#define GCC_INTERNAL_INCLUDE "/usr/lib/gcc/i486-linux-gnu/4.1.2/include"

Ah, it is hard-coding the path.

Well, sort of. The patch is actually determined at sparse build time, so right now sparse requires you to build it against the same compiler that you are currently using. If you rebuild sparse with your shiny new compiler you'll notice that it'll end up with a different path in pre-process.h.

johannes

#2 lore raw
Quoted 6 lines · Johannes Berg

Well, sort of. The patch is actually determined at sparse build time, so right now sparse requires you to build it against the same compiler that you are currently using. If you rebuild sparse with your shiny new compiler you'll notice that it'll end up with a different path in pre-process.h.

Then the short-term solution for debian is clear. Simply rebuild the package and add a dependency on a specific gcc version.

Jörn

Signature · 7 lines
Fantasy is more important than knowledge. Knowledge is limited,
while fantasy embraces the whole world.
-- Albert Einstein
--
To unsubscribe from this list: send the line "unsubscribe linux-sparse" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#8 lore raw
In reply to Jörn Engel

On Tue, 2008-11-11 at 14:58 +0100, Jörn Engel wrote:
[...]

Quoted 2 lines

So what is the right solution to this problem? Call "gcc --version" and parse the (deliberately hard) output? Or make a copy of the gcc headers

There is gcc -dumpversion. But it doesn't help if your compiler is from e.g. /opt/gcc-4.1.2 (where mine usually are).
And which, type -all, and similar also doesn't really help as the reported pathname can be actually the ccache binary ....

So you want to use the output of
---- snip ----
gcc -v -E - </dev/null 2>&1 >/dev/null | sed -n -e '/^#include <\.\.\.> search starts here:/,/^End of search list\./s/^ \(.*\)/\1/p' ---- snip ----
which also seems to work with non-standard paths (like /opt) and ccache
in between.

Bernd

PS: The core of the above line is from
http://sourceware.org/ml/crossgcc/2006-12/msg00038.html. The quite trivial sed filter is by me.

Signature · 9 lines
Firmix Software GmbH                   http://www.firmix.at/
mobil: +43 664 4416156                 fax: +43 1 7890849-55
          Embedded Linux Development and Services


--
To unsubscribe from this list: send the line "unsubscribe linux-sparse" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#3 lore raw

Bug#505177: Hard-coded gcc header path

In reply to Jörn Engel
Quoted 10 lines · Jörn Engel
Quoted 6 lines · Johannes Berg

Well, sort of. The patch is actually determined at sparse build time, so right now sparse requires you to build it against the same compiler that you are currently using. If you rebuild sparse with your shiny new compiler you'll notice that it'll end up with a different path in pre-process.h.

Then the short-term solution for debian is clear. Simply rebuild the package and add a dependency on a specific gcc version.

Which sucks badly because the default gcc is not always the same on every arch, and more importantly, people may want to use different GCCs. It would look like a better idea to ask gcc for its include path properly.

Signature · 3 lines
·O·  Pierre Habouzit
··O                                                [email protected]
OOO                                                http://www.madism.org
#4 lore raw
Quoted 9 lines · Pierre Habouzit
Quoted 3 lines · Jörn Engel

Then the short-term solution for debian is clear. Simply rebuild the package and add a dependency on a specific gcc version.

Which sucks badly because the default gcc is not always the same on every arch, and more importantly, people may want to use different GCCs. It would look like a better idea to ask gcc for its include path properly.

Maybe not even that. It is conceivable to want sparse on systems that don't even have gcc installed. The BSDs seem to favor pcc lately. So the preferred long-term solution would be for sparse to ship its own headers.

Jörn

Signature · 8 lines
To recognize individual spam features you have to try to get into the
mind of the spammer, and frankly I want to spend as little time inside
the minds of spammers as possible.
-- Paul Graham
--
To unsubscribe from this list: send the line "unsubscribe linux-sparse" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#5 lore raw

The problem is not the header files, but that we need to ensure that the proper include path is set up.

This is needed because sparse currently tries to mimic some indeterminate gcc version in terms of predefines. A similar solution would be needed to mimic any other compiler. We obviously need a default path for system headers somehow.

In this case, we could probably get away with accepting a definition of GCC_INTERNAL_INCLUDE on the command line and use that in preference over whatever is in pre-process.h

Morten

#6 lore raw
Quoted 4 lines · Morten Welinder

The problem is not the header files, but that we need to ensure that the proper include path is set up.

Quoted 8 lines

This is needed because sparse currently tries to mimic some indeterminate gcc version in terms of predefines. A similar solution would be needed to mimic any other compiler. We obviously need a default path for system headers somehow.

In this case, we could probably get away with accepting a definition of GCC_INTERNAL_INCLUDE on the command line and use that in preference over whatever is in pre-process.h

That doesn't strike me as an improvement. I have a workaround, which is to create a symlink in /usr/lib/gcc/i486-linux-gnu/ that matches what sparse expects. If I had to set GCC_INTERNAL_INCLUDE on every invocation, I'd consider that worse than my current workaround.

Either sparse gets smart enough to find the headers [1] or it ships its own set. Anything else will continue to fail on one machine or another.

[1] Bernd Petrovitsch sent me this neat one-liner in a private mail:
gcc -v -E - </dev/null 2>&1 >/dev/null | sed -n -e '/^#include <\.\.\.> search starts here:/,/^End of search list\./s/^ \(.*\)/\1/p

We could add that to cgcc to set additional include paths.

Jörn

Signature · 8 lines
Public Domain  - Free as in Beer
General Public - Free as in Speech
BSD License    - Free as in Enterprise
Shared Source  - Free as in "Work will make you..."
--
To unsubscribe from this list: send the line "unsubscribe linux-sparse" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#7 lore raw

On Die, 2008-11-11 at 23:32 +0100, Jörn Engel wrote:
[...]

Quoted 2 lines

Either sparse gets smart enough to find the headers [1] or it ships its own set. Anything else will continue to fail on one machine or another.

I don't think that shipping a copy of .h files from one gcc version will seriously help people using pcc (or any other compiler). But I'm probably missing something here.

Quoted 4 lines

[1] Bernd Petrovitsch sent me this neat one-liner in a private mail:
gcc -v -E - </dev/null 2>&1 >/dev/null | sed -n -e '/^#include <\.\.\.> search starts here:/,/^End of search list\./s/^ \(.*\)/\1/p

We could add that to cgcc to set additional include paths.

And there probably are similar ones for pcc or others.

Bernd

Signature · 8 lines
Firmix Software GmbH                   http://www.firmix.at/
mobil: +43 664 4416156                 fax: +43 1 7890849-55
          Embedded Linux Development and Services

--
To unsubscribe from this list: send the line "unsubscribe linux-sparse" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html