freebsd-skq/gnu/usr.bin/cvs/BUGS
peter 2dbe609ba8 Import CVS-1.6.3-951211.. Basically, this is the cvs-1.6.2 release
plus a couple of minor changes..  

Some highlights of the new stuff that was not in the old version:
 - remote access support.. full checkout/commit/log/etc..
 - much improved dead file support..
 - speed improvements
 - better $CVSROOT handling
 - $Name$ support
 - support for a "cvsadmin" group to cut down rampant use of "cvs admin -o"
 - safer setuid/setgid support
 - many bugs fixed.. :-)
 - probably some new ones.. :-(
 - more that I cannot remember offhand..
1995-12-10 22:31:43 +00:00

249 lines
11 KiB
Plaintext

* `cvs checkout -d nested/dir/path <module>' just doesn't work. The
simpler version -- `cvs checkout -d single-dir <module>' works,
however.
* CVS leaves .#mumble files around when a conflict occurs. (Note:
this is intentional and is documented in doc/cvs.texinfo. Of course
whether it is a good idea is a separate question).
* pcl-cvs doesn't like it when you try to check in a file which isn't
up-to-date. The messages produced by the server perhaps don't match
what pcl-cvs is looking for.
* From: Roland McGrath <roland@gnu.ai.mit.edu>
To: Cyclic CVS Hackers <cyclic-cvs@cyclic.com>
Subject: weird bug
Date: Sat, 25 Mar 1995 16:41:41 -0500
X-Windows: Even your dog won't like it.
I just noticed some droppings on my disk from what must be a pretty weird
bug in remote CVS.
In my home directory on a repository machine I use, I find:
drwxr-xr-x 4 roland staff 512 Mar 7 14:08 cvs-serv28962
drwxr-xr-x 4 roland staff 512 Mar 7 14:11 cvs-serv28978
drwxr-xr-x 4 roland staff 512 Mar 7 15:13 cvs-serv29141
OK, so these are leftover cruft from some cvs run that got aborted.
Well, it should clean up after itself, but so what.
The last one is pretty dull; the real weirdness is the contents of the
first two directories.
duality 77 # ls -RF cvs-serv28978/
CVS/ cvs-serv28978/
cvs-serv28978/CVS:
Entries Repository
cvs-serv28978/cvs-serv28978:
arpa/
cvs-serv28978/cvs-serv28978/arpa:
CVS/ cvs-serv28978/
cvs-serv28978/cvs-serv28978/arpa/CVS:
Entries Repository
cvs-serv28978/cvs-serv28978/arpa/cvs-serv28978:
assert/
cvs-serv28978/cvs-serv28978/arpa/cvs-serv28978/assert:
CVS/ cvs-serv28978/
cvs-serv28978/cvs-serv28978/arpa/cvs-serv28978/assert/CVS:
Entries Repository
cvs-serv28978/cvs-serv28978/arpa/cvs-serv28978/assert/cvs-serv28978:
bare/
cvs-serv28978/cvs-serv28978/arpa/cvs-serv28978/assert/cvs-serv28978/bare:
CVS/ cvs-serv28978/
cvs-serv28978/cvs-serv28978/arpa/cvs-serv28978/assert/cvs-serv28978/bare/CVS:
Entries Repository
cvs-serv28978/cvs-serv28978/arpa/cvs-serv28978/assert/cvs-serv28978/bare/cvs-serv28978:
conf/
cvs-serv28978/cvs-serv28978/arpa/cvs-serv28978/assert/cvs-serv28978/bare/cvs-serv28978/conf:
CVS/ cvs-serv28978/
cvs-serv28978/cvs-serv28978/arpa/cvs-serv28978/assert/cvs-serv28978/bare/cvs-serv28978/conf/CVS:
Entries Repository
cvs-serv28978/cvs-serv28978/arpa/cvs-serv28978/assert/cvs-serv28978/bare/cvs-serv28978/conf/cvs-serv28978:
crypt/
cvs-serv28978/cvs-serv28978/arpa/cvs-serv28978/assert/cvs-serv28978/bare/cvs-serv28978/conf/cvs-serv28978/crypt:
CVS/ cvs-serv28978/
cvs-serv28978/cvs-serv28978/arpa/cvs-serv28978/assert/cvs-serv28978/bare/cvs-serv28978/conf/cvs-serv28978/crypt/CVS:
Entries Repository
cvs-serv28978/cvs-serv28978/arpa/cvs-serv28978/assert/cvs-serv28978/bare/cvs-serv28978/conf/cvs-serv28978/crypt/cvs-serv28978:
csu/
cvs-serv28978/cvs-serv28978/arpa/cvs-serv28978/assert/cvs-serv28978/bare/cvs-serv28978/conf/cvs-serv28978/crypt/cvs-serv28978/csu:
CVS/ cvs-serv28978/
cvs-serv28978/cvs-serv28978/arpa/cvs-serv28978/assert/cvs-serv28978/bare/cvs-serv28978/conf/cvs-serv28978/crypt/cvs-serv28978/csu/CVS:
Entries Repository
cvs-serv28978/cvs-serv28978/arpa/cvs-serv28978/assert/cvs-serv28978/bare/cvs-serv28978/conf/cvs-serv28978/crypt/cvs-serv28978/csu/cvs-serv28978:
ctype/
cvs-serv28978/cvs-serv28978/arpa/cvs-serv28978/assert/cvs-serv28978/bare/cvs-serv28978/conf/cvs-serv28978/crypt/cvs-serv28978/csu/cvs-serv28978/ctype:
CVS/ cvs-serv28978/
[...]
ls: cvs-serv28978/cvs-serv28978/arpa/cvs-serv28978/assert/cvs-serv28978/bare/cvs-serv28978/conf/cvs-serv28978/crypt/cvs-serv28978/csu/cvs-serv28978/ctype/cvs-serv28978/dirent/cvs-serv28978/elf/cvs-serv28978/gnu/cvs-serv28978/gnulib/cvs-serv28978/grp/cvs-serv28978/hurd/cvs-serv28978/hurd/hurd/cvs-serv28978/inet/cvs-serv28978/inet/arpa/cvs-serv28978/inet/netinet[...]/cvs-serv28978/posix/cvs-serv28978/posix/glob/cvs-serv28978/posix/gnu/cvs-serv28978/posix/sys/cvs-serv28978/protocols/cvs-serv28978/pwd/cvs-serv28978/resolv/cvs-serv28978/resolv/arpa/cvs-serv28978/resolv/sys/cvs-serv28978/resource/cvs-serv28978/resource/sys/cvs-serv28978/rpc/cvs-serv28978/setjmp/cvs-serv28978/signal/cvs-serv28978/signal/sys/cvs-serv28978/socket/cvs-serv28978/socket: File name too long
cvs-serv28978/cvs-serv28978/arpa/cvs-serv28978/assert/cvs-serv28978/bare/cvs-serv28978/conf/cvs-serv28978/crypt/cvs-serv28978/csu/cvs-serv28978/ctype/cvs-serv28978/dirent/cvs-serv28978/elf/cvs-serv28978/gnu/cvs-serv28978/gnulib/cvs-serv28978/grp/cvs-serv28978/hurd/cvs-serv28978/hurd/hurd/cvs-serv28978/inet/cvs-serv28978/inet/arpa/cvs-serv28978/inet/netinet[...]/cvs-serv28978/posix/glob/cvs-serv28978/posix/gnu/cvs-serv28978/posix/sys/cvs-serv28978/protocols/cvs-serv28978/pwd/cvs-serv28978/resolv/cvs-serv28978/resolv/arpa/cvs-serv28978/resolv/sys/cvs-serv28978/resource/cvs-serv28978/resource/sys/cvs-serv28978/rpc/cvs-serv28978/setjmp/cvs-serv28978/signal/cvs-serv28978/signal/sys/cvs-serv28978/socket/cvs-serv28978:
* From: billr@mpd.tandem.com (Bill Robertson)
Subject: Problem with rtag and the -D option
Date: Fri, 17 Mar 1995 10:53:29 -0600 (CST)
I have been trying to use the -D option to specify a date for tagging, but
rtag does not recognize the -D option. It is documented to do so and I've
tested the use of -D with cvs update and cvs diff and it works fine there.
* We need some version numbers really badly. Are there some
(and Charles Hannum is just not including them in his reports), or do
we simply have no reliable way to distinguish between the various
versions of rCVS people on the list are running?
Now that I think of it, version numbers present a problem when
people can update their sources anytime and rebuild. I think the
solution is to increment a minor version number *every* time a bug is
fixed, so we can identify uniquely what someone is running when they
submit a report. This implies recording the version increments in the
ChangeLog; that way we can just look to see where a particular version
lies in relation to the flow of changing code.
Should we be doing same with Guppy? I guess not -- it's only
important when you have people who are updating directly from your
development tree, which is the case with the remote-cvs folks.
Thoughts?
* (Charles Hannum <mycroft@ai.mit.edu>) has these bugs:
I just tossed remote CVS at a fairly large source tree that I already
had, and noticed a few problems:
1) server.c assumes that /usr/tmp is a valid default for the place to
store files uploaded from the client. There are a number of systems
that now use /var/tmp. These should probably be detected by autoconf.
2) The server deals rather ungracefully with the tmp directory
becoming full.
3) There's some oddness with relative paths in Repository files that
causes the directory prefix to be added twice; e.g. if I have CVSROOT
set to `machine:/this/dir', and I try to update in a directory whose
Repository file says `src/bin', the server looks in
`/this/dir/machine:/this/dir/src/bin'.
* From: "Charles M. Hannum" <mycroft@ai.mit.edu>
To: jimb@duality.gnu.ai.mit.edu, roland@duality.gnu.ai.mit.edu
Subject: Serious flaw in remote CVS
Date: Wed, 22 Feb 1995 20:54:36 -0500
I just found a major flaw in the current implementation. Because the
sockets are not changed to non-blocking mode, write(2)s can hang. In
some cases, this happens on both sides at the same time, with the
socket buffers full in both directions. This causes a deadlock,
because both processes are stuck in I/O wait and thus never drain
their input queues.
Until this is fixed, I can't use it. I'll look at the problem myself
at some point, but I don't know when.
From: "Charles M. Hannum" <mycroft@ai.mit.edu>
To: remote-cvs@cyclic.com
Cc: jimb@totoro.bio.indiana.edu
Subject: Re: forwarded message from Charles M. Hannum
Date: Wed, 22 Feb 1995 22:07:07 -0500
FYI, this happened because the tmp directory on the server became
full. Somehow the server started interpreting the files the client
was sending as commands, and started spewing tons of errors.
Apparently the errors are sent with blocking I/O, or something, and
thus allowed the deadlock to happen.
* From: "Charles M. Hannum" <mycroft@ai.mit.edu>
To: remote-cvs@cyclic.com
Subject: Regarding that relative path problem
Date: Thu, 23 Feb 1995 02:41:51 -0500
This is actually more serious. If you have `bar.com:/foo' as your CVS
root directory, then:
1) When you check things out, the Repository files will contain
`/foo/...' (i.e. without the machine name), which makes little sense.
2) If you instead have a relative path, when the Repository file is
read, `bar.com:/foo' is prepended. This is sent to the server, but
confuses it, because it's not expecting the machine name to be
prepended.
A slightly klugy fix would be to have the client prepend the machine
name when writing a new Repository file, and strip it off before
sending one to the server. This would be backward-compatible with the
current arrangement.
* From: "Charles M. Hannum" <mycroft@ai.mit.edu>
To: remote-cvs@cyclic.com
Subject: Still one more bug
Date: Sat, 25 Feb 1995 17:01:15 -0500
mycroft@duality [1]; cd /usr/src/lib/libc
mycroft@duality [1]; cvs diff -c2 '-D1 day ago' -Dnow
cvs server: Diffing .
cvs server: Diffing DB
cvs [server aborted]: could not chdir to DB: No such file or directory
mycroft@duality [1];
`DB' is an old directory, which no longer has files in it, and is
removed automatically when I use the `-P' option to checkout.
This error doesn't occur when run locally.
P.S. Is anyone working on fixing these bugs?
* From: Roland McGrath <roland@gnu.ai.mit.edu>
To: Cyclic CVS Hackers <cyclic-cvs@cyclic.com>
Subject: bizarre failure mode
Date: Tue, 7 Mar 95 14:17:28 -0500
This is pretty weird:
CVS_SERVER='TMPDIR=. /usr/local/bin/cvs' ../cvs-build/src/cvs update -q
cvs [server aborted]: could not get working directory: Result too large
[Exit 1]
asylum 29 % grep 'Result too large' /usr/include/sys/errno.h
#define ERANGE 34 /* Result too large */
Now, getcwd fails with ERANGE when the buffer is too small. But I don't
know why that would be the case; I don't think there are exceptionally long
directory names involved. It would be robust to notice ERANGE and use a
bigger buffer. But I suspect something weirder is going on.
The repository in question in duality.gnu.ai.mit.edu:/gd4/gnu/cvsroot/libc.
Send me a PGP-signed message if you want the password to use the machine
where the problem showed up.