temp preprocessed filename

temp preprocessed filename

[Michael Morrell]

I'm tracking down some differences in the object files from ostensibly the same compilation flags of the same input.  I've got two .o files that should be the same, but when I run:

  readelf -wl <object file> | grep ii

to see the debug.line info, I see a difference.  The filename that is shown
is of the form <basename>.tmp,<hostname>.<number>.ii

and <number> is different.  I've looked through the source and can't find
where that filename is generated.  I want to know why <number> would change
and how to prevent it from doing so.

Can anyone point me in the right direction?

Thanks,

  Michael

Re: temp preprocessed filename

[Ian Lance Taylor]

Michael Morrell <morrell@alumni.ucsd.edu> writes:

> I'm tracking down some differences in the object files from ostensibly the same compilation flags of the same input.  I've got two .o files that should be the same, but when I run:
>
>   readelf -wl <object file> | grep ii
>
> to see the debug.line info, I see a difference.  The filename that is shown
> is of the form <basename>.tmp,<hostname>.<number>.ii
>
> and <number> is different.  I've looked through the source and can't find
> where that filename is generated.  I want to know why <number> would change
> and how to prevent it from doing so.
>
> Can anyone point me in the right direction?

I think you can get names of that sort when using distcc and/or
ccache.  Are you using either of those?  Check the exact path of the
compiler.

Ian

Re: temp preprocessed filename

[Michael Morrell]

Thanks for the hint.  I am using ccache.  Perhaps that makes it a ccache issue and not appropriate for this list?

  Michael

--- On Sat, 2/27/10, Ian Lance Taylor <iant@google.com> wrote:

> From: Ian Lance Taylor <iant@google.com>
> Subject: Re: temp preprocessed filename
> To: morrell@alumni.ucsd.edu
> Cc: gcc-help@gcc.gnu.org
> Date: Saturday, February 27, 2010, 2:41 PM
> Michael Morrell <morrell@alumni.ucsd.edu>
> writes:
> 
> > I'm tracking down some differences in the object files
> from ostensibly the same compilation flags of the same
> input.  I've got two .o files that should be the same,
> but when I run:
> >
> >   readelf -wl <object file> |
> grep ii
> >
> > to see the debug.line info, I see a difference. 
> The filename that is shown
> > is of the form
> <basename>.tmp,<hostname>.<number>.ii
> >
> > and <number> is different.  I've looked
> through the source and can't find
> > where that filename is generated.  I want to know
> why <number> would change
> > and how to prevent it from doing so.
> >
> > Can anyone point me in the right direction?
> 
> I think you can get names of that sort when using distcc
> and/or
> ccache.  Are you using either of those?  Check
> the exact path of the
> compiler.
> 
> Ian
>

Re: temp preprocessed filename

[Ian Lance Taylor]

Michael Morrell <morrell@alumni.ucsd.edu> writes:

> Thanks for the hint.  I am using ccache.  Perhaps that makes it a
> ccache issue and not appropriate for this list?

Yes, I suspect that it is a ccache issue.  I don't know enough about
ccache to give any suggestions.

Ian

> --- On Sat, 2/27/10, Ian Lance Taylor <iant@google.com> wrote:
>
>> From: Ian Lance Taylor <iant@google.com>
>> Subject: Re: temp preprocessed filename
>> To: morrell@alumni.ucsd.edu
>> Cc: gcc-help@gcc.gnu.org
>> Date: Saturday, February 27, 2010, 2:41 PM
>> Michael Morrell <morrell@alumni.ucsd.edu>
>> writes:
>> 
>> > I'm tracking down some differences in the object files
>> >From ostensibly the same compilation flags of the same
>> input.  I've got two .o files that should be the same,
>> but when I run:
>> >
>> >   readelf -wl <object file> |
>> grep ii
>> >
>> > to see the debug.line info, I see a difference. 
>> The filename that is shown
>> > is of the form
>> <basename>.tmp,<hostname>.<number>.ii
>> >
>> > and <number> is different.  I've looked
>> through the source and can't find
>> > where that filename is generated.  I want to know
>> why <number> would change
>> > and how to prevent it from doing so.
>> >
>> > Can anyone point me in the right direction?
>> 
>> I think you can get names of that sort when using distcc
>> and/or
>> ccache.  Are you using either of those?  Check
>> the exact path of the
>> compiler.
>> 
>> Ian
>> 

Re: temp preprocessed filename

[Michael Morrell]

Luckily(?), this isn't a ccache issue.  I was able to reproduce it with
g++.  We're using an ARM cross-compiler from CodeSourcery (Sourcery G++
4.3-234).

To reproduce what I'm seeing:

 % cat foo.cpp
 #include <cstdlib>

 void foo() {}
 % g++ -g -E foo.cpp > foo.ii
 % g++ -g -c foo.ii
 % readelf -wl foo.o | grep foo
   1	0	0	0	foo.cpp
   6	0	0	0	foo.ii

There is something in <cstdlib> that is triggering this behavior, but I
don't know if it is a cstdlib bug or a compiler bug.

I'll file a bug report with CodeSourcery, but was curious if anyone here
had any thoughts.  I will also try to dig into <cstdlib> and see if I
can isolate this further.

  Michael

Re: temp preprocessed filename

[Michael Morrell]

Here's a smaller testcase:

  typedef int ptrdiff_t;

  namespace std {
    using ::ptrdiff_t;
  }

  void foo() {}

when you run:

  g++ -g -E foo.cpp > foo.ii; g++ -g -c foo.ii

the foo.o produced contains a reference to foo.ii in .debug_line.

  Michael

Re: temp preprocessed filename

[Cedric Roux]

Michael Morrell wrote:
> Luckily(?), this isn't a ccache issue.  I was able to reproduce it with
> g++.  We're using an ARM cross-compiler from CodeSourcery (Sourcery G++
> 4.3-234).
> 
> To reproduce what I'm seeing:
> 
>  % cat foo.cpp
>  #include <cstdlib>
> 
>  void foo() {}
>  % g++ -g -E foo.cpp > foo.ii
>  % g++ -g -c foo.ii
>  % readelf -wl foo.o | grep foo
>    1	0	0	0	foo.cpp
>    6	0	0	0	foo.ii
> 
> There is something in <cstdlib> that is triggering this behavior, but I
> don't know if it is a cstdlib bug or a compiler bug.
> 
> I'll file a bug report with CodeSourcery, but was curious if anyone here
> had any thoughts.  I will also try to dig into <cstdlib> and see if I
> can isolate this further.
> 
>   Michael

doing:

% cat foo.cpp
using namespace bar;

does include foo.ii.

When you do readelf -wlLiaprmfFsoR foo.o
(let's be a bit barbarian)
in the .debug_info you see that std is seen
as being declared in foo.ii (DW_AT_decl_file   : 2)
(where 2 points to foo.ii) instead of foo.cpp.

Don't know if it's expected behavior but maybe
your problem is this one.

Re: temp preprocessed filename

[Cedric Roux]

> % cat foo.cpp
> using namespace bar;

by the way, if it may help, bypassing
the foo.ii steps and directly doing
a "g++ -g -c foo.cpp" declares this line
(or whatever) to be declared in "<built-in>"