Re: [PATCH 5/6] xfs: centralize error tag definitions

[Christoph Hellwig <hch@lst.de>]

On Mon, Sep 15, 2025 at 12:10:01PM -0700, Darrick J. Wong wrote:
> > -#ifndef __XFS_ERRORTAG_H_
> > +#if !defined(__XFS_ERRORTAG_H_) || defined(XFS_ERRTAG)
> 
> Hrmm.  So now xfs_errortag.h has two uses: one where we just #include
> it without defining XFS_ERRTAG, and we get the XFS_ERRTAG_* values?

Yes.

> and I guess a broken XFS_ERRTAGS that isn't fully defined?

Yes.  I wouldn't call it broken, just not usable :)

> And then you're allowed to #include it again, provided you enclose that
> inside a #define XFS_ERRTAG and a user of an XFS_ERRTAGS symbol, in
> which case it actually spits out a usable XFS_ERRTAGS?

Yes.

> That could be documented better.
> 
> "There are two ways to use this header file.  The first way is to
> #include it bare, which will define all the XFS_ERRTAG_* error injection
> knobs for use with the XFS_TEST_ERROR macro.  The second way is to
> enclose the #include with a #define for an XFS_ERRTAG macro, in which
> case the header will define an XFS_ERRTAGS macro that expands to one
> XFS_ERRTAG use for each defined error injection knob."

I can add that.

> > +
> > +#define XFS_ERRTAGS \
> 
> Do you need to #undef XFS_ERRTAGS here to prevent the W=XXX build and/or
> static tools from whining about redefined macros?

XFS_ERRTAGS is redefined identically, which is allowed.  And we can't
undef it here, because we use the definition in the .c files.  We could
undef it after using it, but that feels a bit pointless.

> 
> Also sorta wondering if for cleanliness this ought to be:
> 
> #ifdef XFS_ERRTAG
> # define XFS_ERRTAGS \
> <giant table>
> #endif /* XFS_ERRTAG */
> 
> To avoid cluttering up the macro namespace if the file #include'ing
> xfs_errortag.h didn't define XFS_ERRTAG and doesn't actually want the
> ERRTAGS expansion?

That does sound doable, let me give it a try.