On Wed, Mar 30, 2011 at 01:35:42PM +0200, Daniel Shahaf wrote:
> > > So you're doing:
> > > 
> > > * get error number
> > > * construct an svn_error_t object (that's includes a few malloc()s)
> > > * check error number
> > > * destroy svn_error_t object
> > > 
> > > Won't it be simpler to avoid constructing the svn_error_t object in the
> > > first place (when errno is EPIPE)?
> > 
> > Yes, that's what I had initially.
> > Something like: if (errno == EPIPE) ...
> > But I wasn't sure if that would need #ifdef on windows.
> > And I didn't want to break the windows bots.
> > So I decided to let APR figure out how to represent EPIPE.
> > 
> 
> Given the way your patch passes errno to apr_status_t-expecting
> functions, I'd say that APR_STATUS_IS_EPIPE(errno) should work.

Fixed in r1086936.

> > Well, if the pipe write error is somewhere within the chain, we'd
> > have to iterate the entire chain. Is there an API for that?
> 
> Actually there is (svn_error_has_cause()), but what I had in mind while
> writing my reply was "Hey, isn't it sufficient to require that
> SVN_ERR_IO_PIPE_WRITE_ERROR be the outer-most code?"

Well, that's what it's requiring. Hence the comments.

> > > > @@ -383,7 +383,15 @@ svn_opt_print_generic_help2(const char *
> > > >    return;
> > > >  
> > > >   print_error:
> > > > -  svn_handle_error2(err, stderr, FALSE, "svn: ");
> > > > +  /* Issue #3014:
> > > > +   * Don't print anything on broken pipes. The pipe was likely
> > > > +   * closed by the process at the other end. We expect that
> > > > +   * process to perform error reporting as necessary.
> > > > +   *
> > > > +   * ### This assumes that there is only one error in a chain for
> > > > +   * ### SVN_ERR_IO_PIPE_WRITE_ERROR. See svn_cmdline_fputs(). */
> > > > +  if (err->apr_err != SVN_ERR_IO_PIPE_WRITE_ERROR)
> > > > +    svn_handle_error2(err, stderr, FALSE, "svn: ");
> > > 
> > > This bit of code (including the comment, the ###, and the specific
> > > condition) is repeated N times below.
> > > 
> > > Abstract it into a private helper macro?
> > 
> > Good idea.

Though as I said on IRC, I can't come up with a good way of doing this. 

Do you want a macro like this? I think that's ugly.

 SVN_ERR_CALL_UNLESS(err->apr_err, SVN_ERR_IO_PIPE_WRITE_ERROR,
                     svn_handle_error2(err, stderr, FALSE, "svn: "));

#define SVN_ERR_CALL_UNLESS(error_code1, error_code2, expr) \
  if ((error_code1) != (error_code2) (expr)

Something that hard-codes error_code2 to EPIPE would also be ugly.

Note that this macro would need to be shared between several commands,
not just 'svn'.

Reply via email to