-Werror enabled by default on GNU/Hurd

-Werror enabled by default on GNU/Hurd

[Andrea Monaco]

When building gdb on GNU/Hurd, -Werror is enabled by default.  I think
that's a bug: that option is only useful in specific instances; enabling
it by default means that the build fails after a plain ./configure, and
you need to pass --disable-werror to make it succeed.

I tried to locate the problem myself, but the build scripts are very
complex.  Could someone please give me some hints?



Thanks,

Andrea Monaco

Re: -Werror enabled by default on GNU/Hurd

[Simon Marchi]

On 2021-12-28 17:16, Andrea Monaco via Gdb wrote:
> 
> When building gdb on GNU/Hurd, -Werror is enabled by default.  I think
> that's a bug: that option is only useful in specific instances; enabling
> it by default means that the build fails after a plain ./configure, and
> you need to pass --disable-werror to make it succeed.
> 
> I tried to locate the problem myself, but the build scripts are very
> complex.  Could someone please give me some hints?
> 
> 
> 
> Thanks,
> 
> Andrea Monaco

What are the errors you see?  We aim for warning-free builds, so the best
fix would be to get rid of these warnings.

Simon

Re: -Werror enabled by default on GNU/Hurd

[Joel Brobecker]

> > When building gdb on GNU/Hurd, -Werror is enabled by default.  I think
> > that's a bug: that option is only useful in specific instances; enabling
> > it by default means that the build fails after a plain ./configure, and
> > you need to pass --disable-werror to make it succeed.
> > 
> > I tried to locate the problem myself, but the build scripts are very
> > complex.  Could someone please give me some hints?
> 
> What are the errors you see?  We aim for warning-free builds, so the best
> fix would be to get rid of these warnings.

Indeed. If you use the development branch, then we want warnings
to trigger an error -- otherwise, those warnings very rarely get
looked at. The situation is different on release branches, where
warnings are made non-fatal.

-- 
Joel

Re: -Werror enabled by default on GNU/Hurd

[Andrew Burgess]

* Andrea Monaco via Gdb <gdb@sourceware.org> [2021-12-28 23:16:40 +0100]:

> 
> When building gdb on GNU/Hurd, -Werror is enabled by default.  I think
> that's a bug: that option is only useful in specific instances; enabling
> it by default means that the build fails after a plain ./configure, and
> you need to pass --disable-werror to make it succeed.
> 
> I tried to locate the problem myself, but the build scripts are very
> complex.  Could someone please give me some hints?

I just posted this patch to the patches list:

  https://sourceware.org/pipermail/gdb-patches/2021-December/184683.html

with this patch I am now able to build GDB on GNU/Hurd (i686).  This
might help you out.

Thanks,
Andrew

Re: -Werror enabled by default on GNU/Hurd

[Andrew Burgess]

* Andrea Monaco via Gdb <gdb@sourceware.org> [2021-12-29 19:06:04 +0100]:

> 
>   > with this patch I am now able to build GDB on GNU/Hurd (i686).  This
>   > might help you out.
> 
> Thanks, that solves a warning.  But the build still fails with this
> output:
> 
> 
>   CXX    msg_U.o
> cc1: warning: command-line option '-Wno-mismatched-tags' is valid for C++/ObjC++ but not for C
> cc1: warning: command-line option '-Wsuggest-override' is valid for C++/ObjC++ but not for C
> cc1: warning: command-line option '-Wdeprecated-copy' is valid for C++/ObjC++ but not for C
> cc1: warning: command-line option '-Wdeprecated-copy-dtor' is valid for C++/ObjC++ but not for C
> cc1: warning: command-line option '-Wredundant-move' is valid for C++/ObjC++ but not for C
> cc1: warning: command-line option '-Wstrict-null-sentinel' is valid for C++/ObjC++ but not for C
> msg_U.c: In function 'msg_set_init_ports':
> msg_U.c:1044:22: error: conversion from 'unsigned int' to 'short unsigned int' changes value from '4294967295' to '65535' [-Werror=overflow]
>  1044 |   /* msgtl_name = */ MACH_MSG_TYPE_POLYMORPHIC,
>       |                      ^~~~~~~~~~~~~~~~~~~~~~~~~
> msg_U.c: In function 'msg_set_dtable':
> msg_U.c:1838:22: error: conversion from 'unsigned int' to 'short unsigned int' changes value from '4294967295' to '65535' [-Werror=overflow]
>  1838 |   /* msgtl_name = */ MACH_MSG_TYPE_POLYMORPHIC,
>       |                      ^~~~~~~~~~~~~~~~~~~~~~~~~
> cc1: all warnings being treated as errors
> Makefile:1684: recipe for target 'msg_U.o' failed
> make: *** [msg_U.o] Error 1

I don't claim to know anything about GNU/Hurd, this was the first time
I ever installed it.

The file msg_U.c is generated using the mig tool, and the .defs files
binutils-gdb/gdb/msg.defs.  The .defs file just includes
/usr/include/hurd/msg.h.  And the makefile fragment that builds the
msg_U.c is found in binutils-gdb/gdb/config/i386/i386gnu.mn.

I notice in the above, that my version of those two lines is slightly
different, like this:

  /* msgtl_name = */      (unsigned short) MACH_MSG_TYPE_POLYMORPHIC,

Which would solve the error you're seeing.  Finally, here's the
version number information for the mig tool I'm using:

  $ mig --version
  i686-gnu-mig (GNU MIG) 1.8+git20200618

Remembering that you probably understand all this far better than me,
I wonder if you need to update your version of mig?

As a last resort if you patch your generated file, adding the cast
that I see, is this enough to finish building GDB?  Clearly this isn't
a long term solution, but it feels like the above warning is a tooling
issue, rather than a problem with the GDB source.

Thanks,
Andrew

Re: -Werror enabled by default on GNU/Hurd

[Andrea Monaco]

  > with this patch I am now able to build GDB on GNU/Hurd (i686).  This
  > might help you out.

Thanks, that solves a warning.  But the build still fails with this
output:


  CXX    msg_U.o
cc1: warning: command-line option '-Wno-mismatched-tags' is valid for C++/ObjC++ but not for C
cc1: warning: command-line option '-Wsuggest-override' is valid for C++/ObjC++ but not for C
cc1: warning: command-line option '-Wdeprecated-copy' is valid for C++/ObjC++ but not for C
cc1: warning: command-line option '-Wdeprecated-copy-dtor' is valid for C++/ObjC++ but not for C
cc1: warning: command-line option '-Wredundant-move' is valid for C++/ObjC++ but not for C
cc1: warning: command-line option '-Wstrict-null-sentinel' is valid for C++/ObjC++ but not for C
msg_U.c: In function 'msg_set_init_ports':
msg_U.c:1044:22: error: conversion from 'unsigned int' to 'short unsigned int' changes value from '4294967295' to '65535' [-Werror=overflow]
 1044 |   /* msgtl_name = */ MACH_MSG_TYPE_POLYMORPHIC,
      |                      ^~~~~~~~~~~~~~~~~~~~~~~~~
msg_U.c: In function 'msg_set_dtable':
msg_U.c:1838:22: error: conversion from 'unsigned int' to 'short unsigned int' changes value from '4294967295' to '65535' [-Werror=overflow]
 1838 |   /* msgtl_name = */ MACH_MSG_TYPE_POLYMORPHIC,
      |                      ^~~~~~~~~~~~~~~~~~~~~~~~~
cc1: all warnings being treated as errors
Makefile:1684: recipe for target 'msg_U.o' failed
make: *** [msg_U.o] Error 1