Re: Why is ENOTSUP == ECANCELED?

[Roland McGrath]

> That someone was you :)

D'oh!  And now you are going to trust me to tell you how it should be?

> I have not done any actual comparison.  Here is a description of all error
> values in that file:
> 
> 1 - 106: Have a comment "@c DO NOT REMOVE"

That comment is just there is make sure noone editting the manual text
removes the "@comment errno NNN" part, which is what matters.

> 107 - 117: XOPEN error codes added by Thomas (ChangeLog.6), missing that
>            comment
> 118: Occurs twice

Ok, then there is just that one to fix.  Changing ECANCELED should break
only libpthread and its users, whereas ENOTSUP has been around longer.
So ECANCELED should change to a new number.

> Furthermore, several ???/<NR> variants as well as ???/??? with Linux error
> codes.  If these have any effect, they need to be fixed as the <NR> collides
> with the numbers 1 - 118 above.

I would guess that in ???/<NR> the <NR> is the Linux/i386 number.  All the
stupid Linux inventions have to be in errno.texi even if they aren't
probably documented, because that's used to generate sysdeps/gnu/errlist.c,
which is common between GNU/Linux and GNU/Hurd configurations.  However,
only those with a proper number in the field after "@comment errno" get
matched by sysdeps/mach/hurd/errnos.awk and go into bits/errno.h on the Hurd.
If any of the extra names are things we want the Hurd to support, they need
to be assigned nonconflicting numbers to replace the "???/whatever" there now.
AFAIK those are all useless cruft we don't want.