Thread: Compiler warning with 'fast' variable

Any idea why I am seeing this warning with the new pg_start_backup()
'fast' flag?
xlog.c:6917: warning: variable `fast' might be clobbered by`longjmp' or `vfork'

The line is in a PG_ENSURE_ERROR_CLEANUP() block.  This is with gcc
version 2.95.3.

--  Bruce Momjian  <bruce@momjian.us>        http://momjian.us EnterpriseDB
http://enterprisedb.com
 + If your life is a hard drive, Christ can be your backup. +

Bruce Momjian <bruce@momjian.us> writes:
> Any idea why I am seeing this warning with the new pg_start_backup()
> 'fast' flag?

>     xlog.c:6917: warning: variable `fast' might be clobbered by
>     `longjmp' or `vfork'

> The line is in a PG_ENSURE_ERROR_CLEANUP() block.  This is with gcc
> version 2.95.3.

That's pretty bizarre --- I don't see it here with gcc 2.95.3,
and there is no reason for such a warning to appear on a variable
that isn't changed during the function.

We could stick a volatile on it but I'd like to find out why this
particular variable seems to need that.
        regards, tom lane

Tom Lane wrote:
> Bruce Momjian <bruce@momjian.us> writes:
> > Any idea why I am seeing this warning with the new pg_start_backup()
> > 'fast' flag?
> 
> >     xlog.c:6917: warning: variable `fast' might be clobbered by
> >     `longjmp' or `vfork'
> 
> > The line is in a PG_ENSURE_ERROR_CLEANUP() block.  This is with gcc
> > version 2.95.3.
> 
> That's pretty bizarre --- I don't see it here with gcc 2.95.3,
> and there is no reason for such a warning to appear on a variable
> that isn't changed during the function.
> 
> We could stick a volatile on it but I'd like to find out why this
> particular variable seems to need that.

You ready for this;  changing 'bool' to 'int' supressed the warning:
int     fast = PG_GETARG_BOOL(1);

Using 'char' returns the warning.  Changing the assignment to 'true'
also fixes it:
bool        fast = true;

This also generates no warning about longjmp:
bool        fast = PG_GETARG_TEXT(1);

No warning here either:
bool        fast = (bool) PG_GETARG_DATUM(0);

This generates the warning:
bool        fast = ((bool) ((bool) (PG_GETARG_DATUM(1)) != 0));

This does not:
bool        fast = (bool) (PG_GETARG_DATUM(1) != 0);

--  Bruce Momjian  <bruce@momjian.us>        http://momjian.us EnterpriseDB
http://enterprisedb.com
 + If your life is a hard drive, Christ can be your backup. +

Bruce Momjian <bruce@momjian.us> writes:
> Tom Lane wrote:
>> We could stick a volatile on it but I'd like to find out why this
>> particular variable seems to need that.

> You ready for this;  changing 'bool' to 'int' supressed the warning:
>     int     fast = PG_GETARG_BOOL(1);

Well, that's a compiler bug :-(.  Probably platform-specific, too,
which is why I don't see it on HPPA.

I think that the above variant is the least ugly of the alternatives
you show as working, and definitely less ugly than plastering volatile
on it.
        regards, tom lane

Tom Lane wrote:
> Bruce Momjian <bruce@momjian.us> writes:
> > Tom Lane wrote:
> >> We could stick a volatile on it but I'd like to find out why this
> >> particular variable seems to need that.
> 
> > You ready for this;  changing 'bool' to 'int' supressed the warning:
> >     int     fast = PG_GETARG_BOOL(1);
> 
> Well, that's a compiler bug :-(.  Probably platform-specific, too,
> which is why I don't see it on HPPA.
> 
> I think that the above variant is the least ugly of the alternatives
> you show as working, and definitely less ugly than plastering volatile
> on it.

Well, let's leave it alone and see if anyone else find it;  I can mask
it on my end.

--  Bruce Momjian  <bruce@momjian.us>        http://momjian.us EnterpriseDB
http://enterprisedb.com
 + If your life is a hard drive, Christ can be your backup. +