[RFC] Send new ARI warnings to gdb_testers mailing list

[RFC] Send new ARI warnings to gdb_testers mailing list

[Pierre Muller]

  To give more echo to new ARI warnings,
I propose to automatically generate
a email to gdb-testers@sourceware.org
each time the result file
ari.source.bug changes.

The change is contained in gdbadmin/ss/update-web-ari

 I would like reactions on:

1) is this a good idea?
2) is the syntax for the non-empty file correct?
3) is gdb-testers mailing list acceptable for this change?

Pierre Muller




Index: update-web-ari
===================================================================
RCS file: /cvs/gdbadmin/ss/update-web-ari,v
retrieving revision 1.141
diff -u -r1.141 update-web-ari
--- update-web-ari      6 Apr 2009 21:27:02 -0000       1.141
+++ update-web-ari      22 Apr 2009 22:03:12 -0000
@@ -157,10 +157,19 @@

 if ${check_source_p} && test -d "${srcdir}"
 then
+    cp -f ${wwwdir}/ari.source.bug ${wwwdir}/ari.source.old
     echo "`date`: Checking source code" 1>&2
     ( cd "${srcdir}" && /bin/sh $HOME/ss/gdb_find.sh "${project}" | \
        xargs /bin/sh $HOME/ss/gdb_ari.sh -Werror -Wall --print-idx
     ) > ${wwwdir}/ari.source.bug
+    difffile=${wwwdir}/ari.source.diff
+
+    diff ${wwwdir}/ari.source.old ${wwwdir}/ari.source.bug > ${difffile}
+    if [ -s ${difffile} ]; then
+       # Send an email to gdb-testers@sourceware.org
+       mutt -s "ARI warning list change `date`" -i ${difffile} \
+           gdb-testers@sourceware.org
+    fi
 fi

Re: [RFC] Send new ARI warnings to gdb_testers mailing list

[Joel Brobecker]

> To give more echo to new ARI warnings, I propose to automatically
> generate a email to gdb-testers@sourceware.org each time the result
> file ari.source.bug changes.

Generally speaking, why not. We could give it a try and see if
it helps.

However, the way it is implemented right now means that we'll also get
an email when we fix things. I think that would be too much. What do
others think?

-- 
Joel

RE: [RFC] Send new ARI warnings to gdb_testers mailing list

[Pierre Muller]

Thanks for the remarks, Joel,

> However, the way it is implemented right now means that we'll also get
> an email when we fix things. I think that would be too much. What do
> others think?
What about the following patch:
it does copy the old ari.source.bug to ari.source.old
then it removes the line number field from each of the two files
(in order to avoid seeing only line number changes due to code insertion or
deletion),
and compares the two resulting files.
  If the generated diff contains a '>' sign, it means that
a new warning has been generated.

  The only drawback is that the line number also does not appear
on the generated email...

  If someone with good awk knowledge can tell me out to
substitute each line of the diff (without the line number)
by the corresponding line of the files with line numbers,
it would make the email better.


Pierre Muller
Pascal language support maintainer for GDB
... still only fluent in pascal language, getting used to C,
but still ignorant in awk :(




$ cvs diff -u update-web-ari
Index: update-web-ari
===================================================================
RCS file: /cvs/gdbadmin/ss/update-web-ari,v
retrieving revision 1.141
diff -u -r1.141 update-web-ari
--- update-web-ari      6 Apr 2009 21:27:02 -0000       1.141
+++ update-web-ari      1 May 2009 09:19:47 -0000
@@ -157,10 +157,21 @@

 if ${check_source_p} && test -d "${srcdir}"
 then
+    cp -f ${wwwdir}/ari.source.bug ${wwwdir}/ari.source.old
     echo "`date`: Checking source code" 1>&2
     ( cd "${srcdir}" && /bin/sh $HOME/ss/gdb_find.sh "${project}" | \
        xargs /bin/sh $HOME/ss/gdb_ari.sh -Werror -Wall --print-idx
     ) > ${wwwdir}/ari.source.bug
+    difffile=${wwwdir}/ari.source.diff
+    sed -e "s/\([^:]*\):\([^:]*\):\(.*\)/\1:0:\3/" ${wwwdir}/ari.source.old
> ${wwwdir}/ari.source.old-pruned
+    sed -e "s/\([^:]*\):\([^:]*\):\(.*\)/\1:0:\3/" ${wwwdir}/ari.source.bug
> ${wwwdir}/ari.source.bug-pruned
+    diff ${wwwdir}/ari.source.old-pruned ${wwwdir}/ari.source.bug-pruned >
${difffile}-in
+    sed -n -e "/^>.*/p" ${difffile}-in > ${difffile}
+    if [ -s ${difffile} ]; then
+       # Send an email to gdb-testers@sourceware.org
+       mutt -s "ARI warning list change `date`" -i ${difffile}-in \
+           gdb-testers@sourceware.org
+    fi
 fi

Re: [RFC] Send new ARI warnings to gdb_testers mailing list

[Joel Brobecker]

> The only drawback is that the line number also does not appear
> on the generated email...

Initially, I thought that this was a pretty significant drawback.
But I tried to come up with a way of isolating the new line, and
I'm not sure whether this is possible or not. The problem is that
the line number associated to an error/warning can change from
day to day, as we modify the source around it.

We could try your current approach, and see if that helps or not...
Why not activating your change, but with yourself as the the only
recipient, as opposed to gdb-testers? That will allow you to determine
the frequency of such messages, as we as how difficult it is to
determine which entry is the new one...

-- 
Joel

RE: [RFC] Send new ARI warnings to gdb_testers mailing list

[Pierre Muller]

I just committed this with my email address
and will inform you of the first results I get.

Pierre

> -----Message d'origine-----
> De : Joel Brobecker [mailto:brobecker@adacore.com]
> Envoyé : Wednesday, May 06, 2009 6:31 PM
> À : Pierre Muller
> Cc : gdb-patches@sourceware.org
> Objet : Re: [RFC] Send new ARI warnings to gdb_testers mailing list
> 
> > The only drawback is that the line number also does not appear
> > on the generated email...
> 
> Initially, I thought that this was a pretty significant drawback.
> But I tried to come up with a way of isolating the new line, and
> I'm not sure whether this is possible or not. The problem is that
> the line number associated to an error/warning can change from
> day to day, as we modify the source around it.
> 
> We could try your current approach, and see if that helps or not...
> Why not activating your change, but with yourself as the the only
> recipient, as opposed to gdb-testers? That will allow you to determine
> the frequency of such messages, as we as how difficult it is to
> determine which entry is the new one...
> 
> --
> Joel

RE: [RFC] Send new ARI warnings to gdb_testers mailing list

[Pierre Muller]

Here is an update on this issue.

 Since I added this feature, I received 6 emails.
2 of them where due to the 6.8 branch
that got additional warnings due to my changes to gdb_ari.sh
1 email was about more use of gdbarch methods, which of course
is not of interest here.

I improved the checks about failure in update-web-ari
1) I removed all gdbarch category warnings, as they are not 
relevant here.
2) I used sed to replace the lines with zero'ed line numbers
by the original lines of the gdb_ari.sh script.

The last email I got is this:

>>>>>>>>>> Email Title
ARI warning list change Tue May 19 01:55:11 UTC 2009
>>>>>>>>>> Start of email body
1049a1046,1047
> gdb/lm32-tdep.c:293: deprecate: write_memory: Replace write_memory() with
regcache_read() et.al.
> gdb/lm32-tdep.c:337: deprecate: read_memory: Replace read_memory() with
regcache_read() et.al.
58,59d57
< gdb/doublest.h:96: deprecated: deprecated extract_floating: Do not use
deprecated extract_floating, see declaration for details 
< gdb/doublest.h:97: deprecated: deprecated store_floating: Do not use
deprecated store_floating, see declaration for details 
106d103
< gdb/ada-lang.c:1695: deprecated: deprecated set_value_type: Do not use
deprecated set_value_type, see declaration for details
163d159
< gdb/sh64-tdep.c:1580: deprecated: deprecated extract_floating: Do not use
deprecated extract_floating, see declaration for details
1600,1601d1597
< gdb/doublest.c:822: deprecated: deprecated extract_floating: Do not use
deprecated extract_floating, see declaration for details 
< gdb/doublest.c:839: deprecated: deprecated store_floating: Do not use
deprecated store_floating, see declaration for details
>>>>>>>>>> End of email

Should I change the email address to gdb-testers?

Pierre Muller
trying to improve ARI stuff.


> -----Message d'origine-----
> De : gdb-patches-owner@sourceware.org [mailto:gdb-patches-
> owner@sourceware.org] De la part de Pierre Muller
> Envoyé : Wednesday, May 06, 2009 6:46 PM
> À : 'Joel Brobecker'
> Cc : gdb-patches@sourceware.org
> Objet : RE: [RFC] Send new ARI warnings to gdb_testers mailing list
> 
> I just committed this with my email address
> and will inform you of the first results I get.
> 
> Pierre
> 
> > -----Message d'origine-----
> > De : Joel Brobecker [mailto:brobecker@adacore.com]
> > Envoyé : Wednesday, May 06, 2009 6:31 PM
> > À : Pierre Muller
> > Cc : gdb-patches@sourceware.org
> > Objet : Re: [RFC] Send new ARI warnings to gdb_testers mailing list
> >
> > > The only drawback is that the line number also does not appear
> > > on the generated email...
> >
> > Initially, I thought that this was a pretty significant drawback.
> > But I tried to come up with a way of isolating the new line, and
> > I'm not sure whether this is possible or not. The problem is that
> > the line number associated to an error/warning can change from
> > day to day, as we modify the source around it.
> >
> > We could try your current approach, and see if that helps or not...
> > Why not activating your change, but with yourself as the the only
> > recipient, as opposed to gdb-testers? That will allow you to
> determine
> > the frequency of such messages, as we as how difficult it is to
> > determine which entry is the new one...
> >
> > --
> > Joel

Re: [RFC] Send new ARI warnings to gdb_testers mailing list

[Joel Brobecker]

FWIW:

> >>>>>>>>>> Email Title
> ARI warning list change Tue May 19 01:55:11 UTC 2009
> >>>>>>>>>> Start of email body
> 1049a1046,1047
> > gdb/lm32-tdep.c:293: deprecate: write_memory: Replace write_memory() with
> regcache_read() et.al.
> > gdb/lm32-tdep.c:337: deprecate: read_memory: Replace read_memory() with
> regcache_read() et.al.
> 58,59d57
> < gdb/doublest.h:96: deprecated: deprecated extract_floating: Do not use
> deprecated extract_floating, see declaration for details 
> < gdb/doublest.h:97: deprecated: deprecated store_floating: Do not use
> deprecated store_floating, see declaration for details 
> 106d103
> < gdb/ada-lang.c:1695: deprecated: deprecated set_value_type: Do not use
> deprecated set_value_type, see declaration for details
> 163d159
> < gdb/sh64-tdep.c:1580: deprecated: deprecated extract_floating: Do not use
> deprecated extract_floating, see declaration for details
> 1600,1601d1597
> < gdb/doublest.c:822: deprecated: deprecated extract_floating: Do not use
> deprecated extract_floating, see declaration for details 
> < gdb/doublest.c:839: deprecated: deprecated store_floating: Do not use
> deprecated store_floating, see declaration for details
> >>>>>>>>>> End of email

Can we just get the lines that start with "> "? As far as I can tell,
we don't really need the "< " lines, since these are good news.
The hunk headers (eg: 1600,1601d1597) are not really useful either.

> Should I change the email address to gdb-testers?

I would personally be OK with that; and I even think that this would
be useful if all we were sending is the new regressions. Having
the line number where the regression appeared is great.

-- 
Joel

RE: [RFC] Send new ARI warnings to gdb_testers mailing list

[Pierre Muller]

  Hi, 
sorry for answering so late on this email.

  Joel, do you still agree on 
moving the email-to address to gdb-testers?


> -----Message d'origine-----
> De : Joel Brobecker [mailto:brobecker@adacore.com]
> Envoyé : Wednesday, May 20, 2009 7:23 AM
> À : Pierre Muller
> Cc : gdb-patches@sourceware.org
> Objet : Re: [RFC] Send new ARI warnings to gdb_testers mailing list
> 
> FWIW:
> 
> > >>>>>>>>>> Email Title
> > ARI warning list change Tue May 19 01:55:11 UTC 2009
> > >>>>>>>>>> Start of email body
> > 1049a1046,1047
> > > gdb/lm32-tdep.c:293: deprecate: write_memory: Replace
> write_memory() with
> > regcache_read() et.al.
> > > gdb/lm32-tdep.c:337: deprecate: read_memory: Replace read_memory()
> with
> > regcache_read() et.al.
> > 58,59d57
> > < gdb/doublest.h:96: deprecated: deprecated extract_floating: Do not
> use
> > deprecated extract_floating, see declaration for details
> > < gdb/doublest.h:97: deprecated: deprecated store_floating: Do not
> use
> > deprecated store_floating, see declaration for details
> > 106d103
> > < gdb/ada-lang.c:1695: deprecated: deprecated set_value_type: Do not
> use
> > deprecated set_value_type, see declaration for details
> > 163d159
> > < gdb/sh64-tdep.c:1580: deprecated: deprecated extract_floating: Do
> not use
> > deprecated extract_floating, see declaration for details
> > 1600,1601d1597
> > < gdb/doublest.c:822: deprecated: deprecated extract_floating: Do not
> use
> > deprecated extract_floating, see declaration for details
> > < gdb/doublest.c:839: deprecated: deprecated store_floating: Do not
> use
> > deprecated store_floating, see declaration for details
> > >>>>>>>>>> End of email
> 
> Can we just get the lines that start with "> "? As far as I can tell,
> we don't really need the "< " lines, since these are good news.
> The hunk headers (eg: 1600,1601d1597) are not really useful either.

  I have to disagree here:
In some cases, the new warning might just be due to
a change of the ordering of the source (for instance if you
move a function). In such a case, you would
be able to understand that it is not a really new
warning, but a displaced by looking at the

> > Should I change the email address to gdb-testers?
> 
> I would personally be OK with that; and I even think that this would
> be useful if all we were sending is the new regressions. Having
> the line number where the regression appeared is great.

Pierre

Re: [RFC] Send new ARI warnings to gdb_testers mailing list

[Joel Brobecker]

> Joel, do you still agree on moving the email-to address to
> gdb-testers?

No real opinion, actually. If you think it's going to be useful,
we can certainly give it a try.

While you're looking at this: update-web-ari sent an email when
I created the release. Can we avoid this email in that case?

Thanks,
-- 
Joel