Re: Error handling

[Kaloian Doganov]

Yavor Doganov <address@hidden> writes:

    This has the disadvantage that $FILE is the exact target or
    prerequisite, while I was using make functions to compute a
    convenient name

You are right, but this can be easily fixed by changing one line in
`validate-html-notify':

-SUBJ="[GNUN Error] $FILE is not valid XHTML"
+SUBJ="[GNUN Error] ${FILE#../../} is not valid XHTML"
     
    We may as well merge this into validate-html and make `validate-html
    --notify' do this thing.

I thought about this, but:

1) I couldn't find a straightforward way to use `mailfail' in
   `validate-html'.  This is mainly because `validate-html' mangles the
   output of `xmllint' and we do not want to mail this output unmangled.
   So we can't just run `xmllint' through `mailfail'.

   Another approach is to throw away `mailfail' and implement ad-hoc
   mailing commands in `validate-html'.  This might be the right thing
   if we do not found `mailfail' useful in other parts of GNUN.  (What
   about PO-files validation?)

2) It wouldn't be so straightforward to call `validate-html' with
   `--notify EMAIL' option from the GNUmakefile.  Since EMAIL must be
   different in different cases, this will add complexity to the
   calculation of $(VALIDATESKIP) variable.  It can be done, no doubt
   about that, but in this case I prefer one short script instead of
   more lines in GNUmakefile.

That's the main reasons I've proposed the patch in it's current form.

    This, as well as the command in the article rules, has the flaw of my 
    initial implementation.  make will consider the target up-to-date, so it 
    will not be rebuilt next time.  The error message might be forgotten, 
    and the invalid .LANG.html can stay online for a fairly long time.

    You have to touch the prerequisite (.LANG.po) to avoid that.

I am puzzled by your comments.  The prerequisite (.LANG.po) is touched
exactly for this reason.  What I am missing?