Re: Conerned about lack of detail in ChangeLog/commit messges

Re: Conerned about lack of detail in ChangeLog/commit messges

[Richard Kenner]

    One of the major reasons for starting the egcs project was that it was
    so extremely difficult to locate the gcc2 source code - something
    which you are suffering from right now.

At the point in the GCC2 project in question, the problem wasn't difficulty
in getting to the sources (they were available via cvs), but that very few
were interested in doing so.

I have heard many motivations for the start of the egcs project, but the one
above is new to me: I was not aware of any legitimate developer or tester
who had problems getting to GCC development sources at any point.

Re: Conerned about lack of detail in ChangeLog/commit messges

[Martin v. Loewis]

> I have heard many motivations for the start of the egcs project, but the one
> above is new to me: I was not aware of any legitimate developer or tester
> who had problems getting to GCC development sources at any point.

It's probably hard, and may be pointless, to dig in the specific
history today - however I'd like to offer a single data point. I know
that I, personally, found it very hard to even find out how gcc is
being maintained all those years. I had been looking a number of times
since '95, and really all I could find were bug reporting addresses -
these were also recommended for submitting patches. Of course, it may
be that I was not looking hard enough, but I definitely got the
impression that gcc was being developed behind closed doors.

Regards,
Martin

Re: Conerned about lack of detail in ChangeLog/commit messges

[Richard Kenner]

    This is ancient history now...  

True, but I do want to address one point.

    We can see your attitude in the phrase `I am not aware of any
    legitimate development or tester'. The question is who decides what is
    legitimate?

I don't recall much controversy over that particular issue.  By "legitimate"
I simply meant to distinguish somebody who actually wanted to work on the GCC
project from somebody who was just looking for "the latest and greatest"
without any intent to actually contribute to the project.

Re: Conerned about lack of detail in ChangeLog/commit messges

[Mike Stump]

> Date: Sat, 17 Jun 00 11:53:35 EDT
> From: kenner@vlsi1.ultra.nyu.edu (Richard Kenner)
> To: martin@loewis.home.cs.tu-berlin.de
> Cc: gcc@gcc.gnu.org

> I have heard many motivations for the start of the egcs project, but
> the one above is new to me: I was not aware of any legitimate
> developer or tester who had problems getting to GCC development
> sources at any point.

This is ancient history now...  but I've heard of it.  There was no
public web page that was widely talked about that told a random tester
(a new tester) that exact instructions for grabbing the source,
exactly where the sources were and so on.  Even if I am wrong on this
point, it proves my point, if I didn't know, how could anyone else
know?  gcc sources are now mirrored on 52 mirrors around the world.
Part of availability, is _high_ availability.  The instructions are
public, and published on a public web site.  This is also rather new.

We can see your attitude in the phrase `I am not aware of any
legitimate development or tester'. The question is who decides what is
legitimate?

See The Cathedral and the Bazaar,
http://www.tuxedo.org/~esr/writings/cathedral-bazaar , for a far more
in-depth treatment of the subject.

One last parting shot, remember, perception isn't your perception, but
rather, everyone else's.

Re: Conerned about lack of detail in ChangeLog/commit messges

[Richard Kenner]

    Of course, it may be that I was not looking hard enough, 

Or that you were looking quietly: a number of people's first involvement
with GCC development started with a posting to a GCC newgroup or bug or
help address.

Re: Conerned about lack of detail in ChangeLog/commit messges

[Marc Espie]

Yeah, gcc doesn't make a sensible utilisation of cvs, does it ?
At least it's somewhat different from most other projects that use it
that I know.

The proposal of generating the cvs logs from the Changelog is somewhat 
backwards.  It would probably be much, much simpler to generate a ChangeLog
out of cvs commit messages. In fact, this is how other projects usually 
do it. Cvs does yield fairly nice commit messages.  That would solve
one specific problem: all files changed would be automatically mentioned
in the commit message. The format can be tweaked afterwards to be as 
compliant with the Changelog entry as one wishes...

In fact, I would very much prefer having the Changelog match the committed
date, not some other somewhat artificial dates. There are lots of other ways
to refer to a patch, the entry in the gcc-patches archive comes to mind.

And David is quite correct that using the ChangeLog to browse through the
history of one file is a pain... One loses most of the features that make
cvs useful.  For instance, `cvs annotate' is almost useless on gcc, 
since it is very hard to go back from gcc revisions to Changelog entry.

I understand that this is somewhat hard to change, but consider it.
There might be a good reason why other projects are actually USING cvs for
day-to-day work, might there not ? :)

Re: Conerned about lack of detail in ChangeLog/commit messges

[David O'Brien]

On Fri, Jun 09, 2000 at 05:13:08PM +0100, Nick Burrett wrote:
> I think it is much easier to look at a ChangeLog than look at the CVS logs
> for individual files.

It might help if the date format for ChangeLog entries were
standardized upon.

> The `cvs log' has to keep on being re-generated, the ChangeLog is
> there, fast to access, and provides a good overview of all files
> changed, rather than just a few.

I love my DSL, what can I say. :-)   I do see your point.  But, I am
usually doing the opposite -- I just want to know about the chagnes to a
single file.  I find that a single file's change can get lost in the
sometimes huge ChangeLog entries.
  
-- 
-- David    (obrien@NUXI.com)

Re: Conerned about lack of detail in ChangeLog/commit messges

[Nick Burrett]

"David O'Brien" <obrien@NUXI.com> writes:

> On Fri, Jun 09, 2000 at 11:30:21AM -0400, David Edelsohn wrote:
> > David> The diff from 1.25 to 1.26 is large enough to explain the change.
> > David> Looking at ``cvs log gengenrtl.c'' I was unable to find any details on
> > David> the "gcc2-ss-010999" branch.
> > 
> > 	What you list above is the CVS entry, not the ChangeLog entry.
> > The entry in ChangeLog.2 says:
> > 
> >         * Merge from gcc2 snapshot Jan 9, 1999.  See FSFChangeLog for
> >         details.
> 
> Sorry, I was under the impression (from what I've seen in other CVS
> commit messages) that the ChangeLog text is in the commit message.  IMHO
> CVS logs are much easier to use to find the notes on a change than the
> ChangeLog's.  
> 
> GCC ChangeLog's do not use a consistent date format, thus it is time
> consuming to find the entry you are looking for as searching in your
> favorite viewer isn't as good as it could be.  Nor does this ChangeLog
> entry contain the name of the modified file as it should.  In this case,
> I could not use any of the CVS message text in searching for the
> ChangeLog entry except for the word "Merge".

This would be easily resolvable if in future the ChangeLog date and e-mail
header were included in the CVS commit messages.  It would make things
much easier to cross-reference since the CVS date is always going to be
slightly different to the ChangeLog date.  I'm not sure if CVSweb would
cope very well with this though.
 
>  
> > Did you check FSFChangeLog (which contains the ChangeLog entries from the
> > gcc2 development
> 
> I did not.  I forgot that there was a FSFChangeLog vs. just ChangeLog.  I
> guess I'm just used to the other projects who's CVS repo I work with.
> There is an emphasis on detail in commit messages to assist people in
> finding out about changes.  I'm surprised the message above didn't at
> least tell people to see the FSFChangeLog (and give the applicable date
> range) for details on the changes.

Perhaps it would have been better to include the FSFChangeLog directly
into the ChangeLog under a single entry in the same way that is done for
older branches.  It's a bit late in the day for that though.

> > The actual various ChangeLog files are the historical repository, not
> > CVS log entries.
> 
> Yes I know.  IMHO now that the world has access to the CVS logs, they
> could be better used now.

I think it is much easier to look at a ChangeLog than look at the CVS logs
for individual files.  The `cvs log' has to keep on being re-generated,
the ChangeLog is there, fast to access, and provides a good overview
of all files changed, rather than just a few.
 

Nick.

Re: Conerned about lack of detail in ChangeLog/commit messges

[David O'Brien]

On Fri, Jun 09, 2000 at 11:30:21AM -0400, David Edelsohn wrote:
> David> The diff from 1.25 to 1.26 is large enough to explain the change.
> David> Looking at ``cvs log gengenrtl.c'' I was unable to find any details on
> David> the "gcc2-ss-010999" branch.
> 
> 	What you list above is the CVS entry, not the ChangeLog entry.
> The entry in ChangeLog.2 says:
> 
>         * Merge from gcc2 snapshot Jan 9, 1999.  See FSFChangeLog for
>         details.

Sorry, I was under the impression (from what I've seen in other CVS
commit messages) that the ChangeLog text is in the commit message.  IMHO
CVS logs are much easier to use to find the notes on a change than the
ChangeLog's.  

GCC ChangeLog's do not use a consistent date format, thus it is time
consuming to find the entry you are looking for as searching in your
favorite viewer isn't as good as it could be.  Nor does this ChangeLog
entry contain the name of the modified file as it should.  In this case,
I could not use any of the CVS message text in searching for the
ChangeLog entry except for the word "Merge".

 
> Did you check FSFChangeLog (which contains the ChangeLog entries from the
> gcc2 development

I did not.  I forgot that there was a FSFChangeLog vs. just ChangeLog.  I
guess I'm just used to the other projects who's CVS repo I work with.
There is an emphasis on detail in commit messages to assist people in
finding out about changes.  I'm surprised the message above didn't at
least tell people to see the FSFChangeLog (and give the applicable date
range) for details on the changes.

> The actual various ChangeLog files are the historical repository, not
> CVS log entries.

Yes I know.  IMHO now that the world has access to the CVS logs, they
could be better used now.

-- 
-- David    (obrien@NUXI.com)

Re: Conerned about lack of detail in ChangeLog/commit messges

[David Edelsohn]

>>>>> "David O'Brien" writes:

David> The ChangeLog entry for this change is:

David> revision 1.26
David> date: 1999/09/07 05:47:55;  author: law;  state: Exp;  lines: +142 -130
David> Merge in gcc2-ss-010999

David> This lack of detail was very annoying as I spent a lot more time than
David> needed to verify that the differences I noticed in gen* were intentional.
David> The diff from 1.25 to 1.26 is large enough to explain the change.
David> Looking at ``cvs log gengenrtl.c'' I was unable to find any details on
David> the "gcc2-ss-010999" branch.

	What you list above is the CVS entry, not the ChangeLog entry.
The entry in ChangeLog.2 says:

        * Merge from gcc2 snapshot Jan 9, 1999.  See FSFChangeLog for
        details.

Did you check FSFChangeLog (which contains the ChangeLog entries from the
gcc2 development which was occurring in parallel with EGCS?  The actual
various ChangeLog files are the historical repository, not CVS log
entries.

David

Re: Conerned about lack of detail in ChangeLog/commit messges

[Martin v. Loewis]

> The ChangeLog entry for this change is:
> 
>     revision 1.26
>     date: 1999/09/07 05:47:55;  author: law;  state: Exp;  lines: +142 -130
>     Merge in gcc2-ss-010999
> 
> This lack of detail was very annoying as I spent a lot more time than
> needed to verify that the differences I noticed in gen* were intentional.
> The diff from 1.25 to 1.26 is large enough to explain the change.
> Looking at ``cvs log gengenrtl.c'' I was unable to find any details on
> the "gcc2-ss-010999" branch.

Isn't the log message self-explanatory, at least to informed gcc
contributors? That is a merge from gcc2, which used to a project
different from the one that provided the current CVS archives.

One of the major reasons for starting the egcs project was that it was
so extremely difficult to locate the gcc2 source code - something
which you are suffering from right now. The problem has been corrected
today, but can't be corrected retroactively (sp?).

Regards,
Martin

Conerned about lack of detail in ChangeLog/commit messges

[David O'Brien]

Rev 1.26 of gengenrtl.c changed the gen* utils from writing to a file to
writing to stdout.  I noticed this when upgrading the FreeBSD stock
compiler from 2.95.2 to 2.96 and all of a sudden my bmake Makefile broke.

The ChangeLog entry for this change is:

    revision 1.26
    date: 1999/09/07 05:47:55;  author: law;  state: Exp;  lines: +142 -130
    Merge in gcc2-ss-010999

This lack of detail was very annoying as I spent a lot more time than
needed to verify that the differences I noticed in gen* were intentional.
The diff from 1.25 to 1.26 is large enough to explain the change.
Looking at ``cvs log gengenrtl.c'' I was unable to find any details on
the "gcc2-ss-010999" branch.

-- 
-- David    (obrien@NUXI.com)