Peter Maydell [Sun, 6 Nov 2005 21:54:40 +0000 (21:54 +0000)]
Fix various buffer overruns in fmt_scan.c; the bulk of this is passing
buffer length through to decode_rfc2047() and having that function do
sufficient bookkeeping to avoid running off the end of the buffer.
Jon Steinhart [Fri, 19 Nov 2004 05:06:16 +0000 (05:06 +0000)]
Added mail directory argument to folder_addmsg in order to make it possible
to provide a path to the ext_hook call that is mailpath-based. A problem
existed when a folder was a symbolic link and the pwd call would return
the path relative to the filesystem, not to mailpath. A new argument was
needed because there was otherwise no reasonable way to get that path.
Jon Steinhart [Tue, 16 Nov 2004 18:14:34 +0000 (18:14 +0000)]
Fixed problem where the refile hook was being called after a message was
renamed so that it wasn't around for the hook. The hook is now called
before the message file is renamed.
Jon Steinhart [Tue, 12 Oct 2004 20:40:25 +0000 (20:40 +0000)]
Fixed strange bug that prevented a lock from ever being obtained if getting
it failed the first time. The problem was that the string of XXXXXX that
is required by mkstemp() was overwritten the first time through, and so
all subsequent times failed because mkstemp() failed. The fix
reinitializes the tmp file string.
[bug #4302] errno is not always an extern int
The errno declaration is best left to errno.h; several systems use
#define errno (*(___errno()))
or similar
[bug #4296] PUTD() macro can't hack 10 digit values
Once the number exceeded 1000000000, the conversion would fail. An execution of
dp -format "%(clock{text})" "Tue, 01 Jul 2003 21:38:05 +0700" would erroneously print 749660476, instead of 1057070285.
The macros PUTD() and PUTDF() have been replaced by snprintf() calls; These macros
should never have been written in the first place, when system functions exist to do the
work.
[bug #4296] PUTD() macro can't hack 10 digit values
Once the number exceeded 1000000000, the conversion would fail. An execution of
dp -format "%(clock{text})" "Tue, 01 Jul 2003 21:38:05 +0700" would erroneously print 749660476, instead of 1057070285.
The macros PUTD() and PUTDF() have been replaced by snprintf() calls; These macros
should never have been written in the first place, when system functions exist to do the
work.
fix handling of "-list" and "-seq" options, so behavior is orthoganal.
Previously, "pick -list -seq seqname" meant "pick -nolist -seq seqname".
Now "pick -list -seq seqname" and "pick -seq seqname -list" do the same thing.
Thanks to Robert Elz.
update 'mp->lowmsg' from 0 to 1 during regular 'inc' from mail spool;
avoids failure of 'folder_realloc()' for case of no messages in
folder, and more than 100 messages to incorporate. Initially,
'folder_read()' will initialize 'mp->lowmsg' to 0, and 'mp->lowoff'
to 1. So the call would be folder_realloc(mp, 1, 200), and the
comparison 'if (mp->nummsg > 0 && lo > mp->lowmsg)' will fail.
make 'spost' handling of "masquerade:" line of mts.conf containing
'draft_from' same as for 'post'. Debian bug report 144098, filed by
Terran Melconian.
status of action wasn't saved for use by 'N' result, so when 'N' was
used slocal could perform the action even if the previous
action failed. Debian bug report #149745, filed by Daniel Müller.
Apply bug fix reported to Debian (#136976) by Peter Maydell; allows
messages with "Content-Transfer-Encoding: binary" to be
displayed, per RFC 2045 section 6.2 paragraph 4
The first ``inc'' into an empty folder would not update the unseen
sequence. This is because seq_list() will return NULL if mp->nummsg
is zero. Insure that mp->nummsg is incremented.
Jeffrey C Honig [Mon, 11 Aug 2003 01:20:52 +0000 (01:20 +0000)]
When compiling format strings, nmh attempts to avoid multiple parsing
of address and date fields by only inserting calls to the parse
functions (FT_PARSEADDR and FT_PARSEDATE) for a given component once.
The problem with this method is that the initial invocation may
actually be on a code path that is conditionally executed. This can
result cached copies of data from the fields in previous messages to
be used.
My solution is to move this optimization from compile time to run time.
Address and Date parsing calls (FT_PARSEADDR and FT_PARSEDATE) will
always be included. Run time flags are used to prevent these functions
from being run more than once per component per message.
The c_flags field has being converted from a boolean to a bit-field to
facilitate maintenance of the new CT_PARSED value. The result value
that used to be in this field is now the bit CF_TRUE and the
overloaded use of this field by scan() is now the CT_DATEFAB bit.
Some unneeded flags (CT_ADDRPARSE, CT_MYMBOX) have also been removed.
Glenn Burkhardt [Mon, 30 Jun 2003 18:58:07 +0000 (18:58 +0000)]
fix bug #1393, "sortm core dumps", using patch provided by
Robert Elz, kre@munnari.oz.au. The core dump occurred
when "Subject:" was immediately followed by an
alphanumeric, and "-textfield subject" were used
as arguments.
Glenn Burkhardt [Thu, 26 Jun 2003 01:30:36 +0000 (01:30 +0000)]
fix for bug #578 repl leaks umask; there are several other
places in the code where the umask value is not restored,
so there might be other similar bugs
Glenn Burkhardt [Sun, 22 Jun 2003 23:58:07 +0000 (23:58 +0000)]
Allowed gdbm as substitute for Berkeley db1/ndbm.h libdb1 (check for
gdbm/ndbm.h, libgdbm now); send only stdout to /dev/null when checking for
'broken vi', this allows 'vim' on Solaris to work, since 'vim' closes stdin
and uses stderr for input when running a script.
Jon Steinhart [Mon, 18 Nov 2002 18:25:00 +0000 (18:25 +0000)]
The file README-ATTACHMENTS was removed - it has been moved to the docs
subdirectory. I thought that this had already been done but I guess not.
It's unrelated to the rest of these changes.
The remainder of these changes affect the way that nmh is installed. Nmh
has historically prompted the user for installation when any nmh command
is invoked if it doesn't think that it's installed. This is a problem for
programs that want to, for example, interact with nmh programs via pipes.
These changes alter this behavior. Now, when an nmh program thinks that
nmh isn't installed, it tells the user to run install-mh to install it.
A -check option has been added to install-mh that allows the existence of
an installation to be silently checked and indicated via the exit status.
The bulk of this change is in sbr/read_context.c. In the process of making
these changes, I noticed a fair amount of useless code that I removed. In
particular:
1. The program started with a test for whether or not defpath had already
been set. This could only happen if context_read() was called more
than once, which it isn't. This check also existed and was removed
from the following places:
sbr/context_del.c
sbr/context_find.c
sbr/context_replace.c
sbr/seq_read.c
sbr/seq_save.c
uip/flist.c
uip/folder.c
uip/rmf.c
2. A similar test for mypath was removed. No need to test it since it
doesn't happen. I'm just not a believer that
a = 1;
if (a != 1)
error();
is a good way to write code!
3. I removed the code makes a copy of the $HOME environment variable as
returned by getenv(). It's never changed, so it doesn't need copying.
4. I removed code that copied the pw_dir member of the passwd structure
returned by getpwuid() if $HOME isn't set. Yes, the returned structure
is static, but since the function is never called again the returned
values stay valid and don't need copying.
5. I removed the test for a NULL pw_dir member in the passwd structure
returned by getpwuid() since that never happens.
6. I removed code that removed a trailing / from mypath if mypath is more
than one character long. Unnecessary as // is interpreted as / anyway.
mypath never shows up in an error message, so nobody will ever see it.
Besides, this code didn't handle the case of multiple trailing slashes.
7. I removed code that replaced the MH environment variable contents with
an absolute path if it was a relative one. I'm a bit nervous about this
one, only testing will tell. Any nmh program that's invoked will use
the same context_read() code to convert a relative path to an absolute
one anyway. And there's no guarantee (or discussion) of this feature
in any of the documentation.
8. Althout context_read() tested the MH environment variable, install-mh
didn't. This means that the behavior was not consistent between
install-mh and all other nmh programs. I changed install-mh to be
consistent with context_read().
9. install-mh has been installed in the lib directory, not bin. So a
user would be unlikely to find it when prompted to run it. The
installation has been changed to add a link from bin to lib.
projects / mmh / log