Switch rcs_{get,set}_mtime() from returning and taking a time_t to
doing so with a struct timespec and then use tv_nsec = UTIME_OMIT
instead of a (time_t)-1 as a "do nothing" value. They can then
fully preserve the timestamp

ok millert@

When system calls indicate an error they return -1, not some arbitrary
value < 0. errno is only updated in this case. Change all (most?)
callers of syscalls to follow this better, and let's see if this strictness
helps us in the future.

There was an extra newline in some cases and missing one in others.
OK joris@

Delay keyword expansion in co until after the locks have been resolved.

This way the expansion can happen with the correct values and files do not
show up as modified.

kill const and string artithmetic related warnings; ok millert@

Style fixes; from Ilya Kaliman

Convert xfree to free. From Fritjof Bornebusch. ok deraadt

Replace <sys/param.h> with <limits.h> and other less dirty headers where
possible. Annotate <sys/param.h> lines with their current reasons. Switch
to PATH_MAX, NGROUPS_MAX, HOST_NAME_MAX+1, LOGIN_NAME_MAX, etc. Change
MIN() and MAX() to local definitions of MINIMUM() and MAXIMUM() where
sensible to avoid pulling in the pollution. These are the files confirmed
through binary verification.
ok guenther, millert, doug (helped with the verification protocol)

Zap some stored values never read; From Fritjof Bornebusch; ok nicm@

exit in usage functions themselves; from Fritjof Bornebusch; ok nicm@

remove wrong casts; ok millert

add missing header needed by futimes()

ok stsp@

Remove the need for rp_file in parser structure, instead keep only one
FILE pointer in RCSFILE. This fixes some ugliness in closing an
fdopen()ed FILE and its underlying file descriptor.

Notified by Joerg Sonnenberger <joerg at britannica dot bec to de>

discussed with and ok nicm

Set rcs_suffixes to default value on initialization.

ok ray zinovik

Check date_parse return values.

OK xsa

Make BUFs autoextend by default. This was already done in OpenCVS
a while ago.

OK zinovik nicm

Reduce variable/function name and whitespace differences between
cvs/rcs.

OK xsa zinovik

Checking if a file is ASCII should only be done when diffing, because
the non-ASCII characters could be printed to the screen. For
checking in files, checking out files, merging files, and removing
revisions of files, we do so regardless of whether the files are
ASCII or binary.

Fixes PR6031.

OK joris and deraadt.

Rename rcs_diffreg() to diffreg().

OK joris@

Sync revisions and time buffers size to be consistent with each others.
Simplifies further size tweaks if needed.
OK ray@.

If strict locking is disabled, make file writable by owner on checkout.
Matches GNU's behaviour. From Pierre Riteau (long time ago).

Correct exit value when one of the files specified cannot be checked out.
Matches GNU's behaviour. From Pierre Riteau (long time ago).

As done in OpenCVS, general includes cleanup sweep. OK otto@.

If a ,suffix file is given as an arg to ci and co, strip it. Avoids
potential disasters. Initial diff from niallo@, ok niallo@ joris@

make sure checkout_rev() sets correct file permissions via file->rf_mode
in case where fstat() of file->rf_fd won't give us right mode.

fixes PR #5383

ok xsa@

keep correct permissions on the RCS file when locking/unlocking the file
with ci and co. fixes PR 5379.
OK millert@ joris@.

When checking to see if a checked-out copy of the file is the same
as what is in the repo, expand keywords before doing the diff.
OK xsa@

Unlike GNU rcs, our co will not overwrite an existing file, even
if the existing file is not writable. While safer, this can be
annoying. I've changed things to compare the version of the file
being checked out with its current contents and, if they are the
same, to allow the checkout without user intervention. The behavior
when there is a writable version of the file is unchanged. OK xsa@

Simplify stripping of write bits from file mode.
Add support for reusing the checkin message for multiple files, ala GNU
Fix the message when you abort a checkout and the file was not writable.
OK joris@ niallo@

- convert some warnx() to warn(), which fixes a few stupidly un-informative error messages (found by tom@)
- make rcs_choosefile() save wrt errno, input from ray@

ok joris@ xsa@

More strlcpy/strlcat -> snprintf/xasprintf cleanup.

OK joris@

Reduce rcs_buf_release() usage, improving binary support.

OK niallo@

- correctly handle binary files; say bye bye to using c strings for deltatexts.

"slap it in" joris@

Remove a lot of xstrdup() calls in getopt() loops, which are usually
unnecessary. These xstrdup() calls don't call xfree() before anyway,
so if a flag is given multiple times memory leaks would have resulted.

OK joris@

Rename RCSFILE.fd to RCSFILE.rf_fd, plus minor spacing nits. No
binary change.

OK xsa@

Remove rcs_statfile(). It was just a wrapper for rcs_choosefile(),
handling exceptions oddly.

OK joris@

fix printf's

fix informative msg's wrt to -q and stderr. now matches gnu/usr.bin/rcs.

Remove /* NOTREACHED */ comments, now that lint is smart about
__dead. I never should have put them there.

OK xsa@.

o GNU ci returns 0 when a file is reverted (co -l file; ci file).
o Don't unlock file after revert.
o Fix spelling.

Passes extra tests in GNU rcstest.

OK niallo@.

Check return values for all strlcpy, and strlcat calls.

OK xsa@ and probably others.

prevent file races by obtaining an fd for the RCS file and
do our operations on that, this is safe and guarantees we
can operate on the file until we close(2) it.

a fix is coming for the remaining races in our diff code.
okay niallo@ and ray@

fork our code we shared between openrcs/cvs into the openrcs dir.

this was starting to become inhuman to maintain without
ugly ugly hacks in the shared code, and it will be easier
to make specific changes for openrcs without touching the
soon-to-be-replaced opencvs code.

fatal() -> err()/errx() as we try to not depend on heavy cvs_log().
OK joriski.

be consistent in error messages for wrongly specified keyword substitution
mode for `-k'.

move shared functions into rcsutil.[ch]; this makes rcsprog.c cleaner;
"the voices in my head say OK!" joris@.

remove overkill cvs_log() and rather use warn()/warnx().
OK ray@ and discussed with joris@.

switch -p handling to bitwise flags; OK ray@.

switch -q handling to bitwise flags; OK ray@ niallo@.

Plug some memory leaks; ok niallo@

excessive addition to brackets is a mental disease. you will go
blind, too. seek help. no binary change, ok joris

correctly handle RCS files without any revisions;
original diff from ray@, crafted into shape by myself.

okay ray@

*** empty log message ***

spaces

Clean up <rev> handling. Whenever a revision is specified after a
flag, it calls one of two new functions: rcs_setrevstr() or
rcs_setrevstr2(). rcs_setrevstr() sets a string to another string,
and complains if it was set more than once. rcs_setrevstr2() takes
two strings, sets one after the other, and fatal()s if more than
two strings were given.

All <rev> handling is now done in the loop that goes through each
argv. This is necessary for parsing symbols, which will be much
easier after this.

Along the way a lot of memory leaks were cleaned up. There is one
area where rcs_set_rev() is called, which allocates a RCSNUM and
stores it in pb.newrev, but it segfaults whenever I try to rcsnum_free()
it. I put an /* XXX */ comment there for now.

Passes regression tests and the code is less complicated in some
ways (to me).

Suggestions and OK xsa@

change rcs_kwexp_set() return type to void and simplify its use. OK niallo@.

- don't print out warnings about locking when they are actually wrong and
just confuse the user.

ok and input joris@

-z can take no argument.

first part of supporting branches in openrcs. right now we can only
check them out. commit is not working yet, but will be soon.

tested by myself and ray@
okay ray@

More closely matches GNU behavior, passes more GNU rcstest tests.

OK niallo@

- properly implement GNU file modes. basically, checkout will inherit
permissions from rcs file and initial checkin will inherit permissions from
working file.

problem spotted by uwe@

ok ray@ xsa@

Allow the -x flag to be used without an extension.

``that makes sense'' xsa@

- co usage() says we support -I, so set the flags and revision.

ok xsa@

- checkout_rev() can fail, in which case we want a non-zero exit code.

ok joris@

error message fix; respect -q too. Ok ray@.

Follow exit() with /* NOTREACHED*/ for lint.

``OK niallo@'' niallo@

Currently co(1) overwrites files that the current user doesn't have
write permissions for. If you own the directory, that means your
existing file, mode 000, is blown away without prompting you. GNU
RCS does this as well.

This fixes changes co(1) to always prompt if a file exists. This
breaks GNU compatibility but I think it's more important that we
prevent co(1) from overwriting files with strict permissions without
even prompting.

``makes sense'' xsa to an earlier diff

- use shared cvs_yesno() for yesno prompt, fixing an issue spotted by
deraadt@

- while here, remove unecessary cast also spotted by deraadt@

ok ray@ xsa@

-z support for RCS;

missing informative message after call to checkout_rev().

should also have removed those newlines in previous commit.

missing newline in checkout_rev(); spotted by joris@.

our deltatexts were sometimes different from gnu. rcs_getrev() was
unconditionally doing keyword expansion, even when this wasn't what we
wanted e.g. in checkout_rev() and cvs_checkout_rev(). so:

- change semantics of rcs_getrev() to not do keyword expansion
- add an rcs.c api function rcs_kwexp_buf() which does this
- change both checkout_rev() and cvs_checkout_rev() to use this function.

eyeballed by xsa, joris and Ray Lai

remove extra printf;

- fix behaviour when using -u and -l at the same time
- do not display some messages when checking in a file for the first time
OK niallo@.

cleanup, move log.h and rcs.h inclusion into rcsprog.h;

#include's cleanup; ok joris@ niallo@.

- implement lazy-parsing of rcs files, that is only parse as much as we
need. this can save us much work, particularly with very large rcs files.

first of a few important performance improvements.

ok joris@

support co -d, which checks out the first revision who's
date is less than or equal to the given date.

use fatal() if getlogin() fails;

switch to xmalloc stuff, me and xsa@ agreed on this a long
time ago, but we were being held back by jfb. too bad for him.

next step is to use fatal() through out the code for unrecoverable
errors instead of trying to be all nice and fluffy and reach main() again.

ok niallo@ and xsa@

if we are removing a lock that's not present, just check
out the file without complaining;

unused vars, my bad;

complete and correct rcs locking functionality,
it was only done partially and as a bonus, completely wrong.

seriously guys what was up with that?

correctly handle `-wfoo' and `-sbar' errors; OK joris@.

`-mmsg' does not belong here;

check for ownership of already present writable file when checking out;

writeable -> writable

consistency in variables naming;

some more bits for `-w[user]' support;

add support for `-T' and enable it for co(1);
tested and OK niallo@. "Looks good" joris@.

memset();

cannot combine -kv and -l;

initial bits for -T support;

add support for `-kmode';

sync SYNOPSIS with Reality and usage() accordingly;

make `-p' work again as expected;

enable `-xsuffixes' support as it has been added a few days ago
in rcs_statfile();
niallo ok

nuke trailing whitespaces;

- update comment for checkout_rev() for bitwise flags.

- sync checkout_usage();

- implement `-M' option.

- update the comment for checkout_state(); it did not tell the truth
about values returned.

changes from Venice:

- instead of using lots of individual flag variables, use a single int and
bitwise operations on it. while this saves memory, really it saves code
space and reduces complexity. checkout_rev() and checkout_state() in
particular benefit from these changes since their parameters where growing
really long.

- implement `-s' option in co

- implement `-M' option in ci

"I like this, go for it" joris@

nuke trailing whitespaces;

fix output if -p is specified; joris ok

- make a bunch of changes to how we handle verbose output. this brings us
much closer in line with GNU RCS and cleans things up in general.

ok joris@, "looks fine to me" xsa@

add co -p support;
input from xsa@ and niallo@;

use rcs_set_rev() instead of duplicating code all over the
place to set a variable to the given revision number;

support -f flag for co;
'fine' niallo@

spacing;

- sync checkout_usage();

- if revision passed to checkout_rev() is > HEAD, assume HEAD
(like GNU RCS).
- move verbose output into checkout_rev() function itself.

"looks sane" joris@

modularise checkout code into checkout_rev(). this shaves off a number
of lines in ci.c due to code re-use, and will simplify features which are in
the pipeline. no functional changes.

ok joris@

To be fully compatibly with the GNU RCS tools we need to have the
same way of parsing commandline options. Since getopt(3) allows spaces
between arguments and GNU RCS tools does not we needed to roll out our
own way of option handling, and here it is.

ok niallo@

minor knf;

various usage cleanup; ok joris

getlogin(2) sets errno;

sync checkout_usage();

- set permissions on checked out files correctly
- support bare `-u' and `-l' options which will check out HEAD.

note that with OpenRCS you must do `co -l -- <file>' and `co -u -- <file>'
when using bare arguments because our getopt(3) is POSIX compliant.

ok joris@

sync usage and add V flag;

- stop gcc's yapping;
- better output when locking and unlocking revisions;

- better output
- free frev when done

- support -l and -u flags;
- better output;

support -q flag in 'co' and 'rcs'

fix usage

put it basic checkout code

When system calls indicate an error they return -1, not some arbitrary
value < 0. errno is only updated in this case. Change all (most?)
callers of syscalls to follow this better, and let's see if this strictness
helps us in the future.

There was an extra newline in some cases and missing one in others.
OK joris@

Delay keyword expansion in co until after the locks have been resolved.

This way the expansion can happen with the correct values and files do not
show up as modified.

kill const and string artithmetic related warnings; ok millert@

Style fixes; from Ilya Kaliman

Convert xfree to free. From Fritjof Bornebusch. ok deraadt

Replace <sys/param.h> with <limits.h> and other less dirty headers where
possible. Annotate <sys/param.h> lines with their current reasons. Switch
to PATH_MAX, NGROUPS_MAX, HOST_NAME_MAX+1, LOGIN_NAME_MAX, etc. Change
MIN() and MAX() to local definitions of MINIMUM() and MAXIMUM() where
sensible to avoid pulling in the pollution. These are the files confirmed
through binary verification.
ok guenther, millert, doug (helped with the verification protocol)

Zap some stored values never read; From Fritjof Bornebusch; ok nicm@

exit in usage functions themselves; from Fritjof Bornebusch; ok nicm@

remove wrong casts; ok millert

add missing header needed by futimes()

ok stsp@

Remove the need for rp_file in parser structure, instead keep only one
FILE pointer in RCSFILE. This fixes some ugliness in closing an
fdopen()ed FILE and its underlying file descriptor.

Notified by Joerg Sonnenberger <joerg at britannica dot bec to de>

discussed with and ok nicm

Set rcs_suffixes to default value on initialization.

ok ray zinovik

Check date_parse return values.

OK xsa

Make BUFs autoextend by default. This was already done in OpenCVS
a while ago.

OK zinovik nicm

Reduce variable/function name and whitespace differences between
cvs/rcs.

OK xsa zinovik

Checking if a file is ASCII should only be done when diffing, because
the non-ASCII characters could be printed to the screen. For
checking in files, checking out files, merging files, and removing
revisions of files, we do so regardless of whether the files are
ASCII or binary.

Fixes PR6031.

OK joris and deraadt.

Rename rcs_diffreg() to diffreg().

OK joris@

Sync revisions and time buffers size to be consistent with each others.
Simplifies further size tweaks if needed.
OK ray@.

If strict locking is disabled, make file writable by owner on checkout.
Matches GNU's behaviour. From Pierre Riteau (long time ago).

Correct exit value when one of the files specified cannot be checked out.
Matches GNU's behaviour. From Pierre Riteau (long time ago).

As done in OpenCVS, general includes cleanup sweep. OK otto@.

If a ,suffix file is given as an arg to ci and co, strip it. Avoids
potential disasters. Initial diff from niallo@, ok niallo@ joris@

make sure checkout_rev() sets correct file permissions via file->rf_mode
in case where fstat() of file->rf_fd won't give us right mode.

fixes PR #5383

ok xsa@

keep correct permissions on the RCS file when locking/unlocking the file
with ci and co. fixes PR 5379.
OK millert@ joris@.

When checking to see if a checked-out copy of the file is the same
as what is in the repo, expand keywords before doing the diff.
OK xsa@

Unlike GNU rcs, our co will not overwrite an existing file, even
if the existing file is not writable. While safer, this can be
annoying. I've changed things to compare the version of the file
being checked out with its current contents and, if they are the
same, to allow the checkout without user intervention. The behavior
when there is a writable version of the file is unchanged. OK xsa@

Simplify stripping of write bits from file mode.
Add support for reusing the checkin message for multiple files, ala GNU
Fix the message when you abort a checkout and the file was not writable.
OK joris@ niallo@

- convert some warnx() to warn(), which fixes a few stupidly un-informative error messages (found by tom@)
- make rcs_choosefile() save wrt errno, input from ray@

ok joris@ xsa@

More strlcpy/strlcat -> snprintf/xasprintf cleanup.

OK joris@

Reduce rcs_buf_release() usage, improving binary support.

OK niallo@

- correctly handle binary files; say bye bye to using c strings for deltatexts.

"slap it in" joris@

Remove a lot of xstrdup() calls in getopt() loops, which are usually
unnecessary. These xstrdup() calls don't call xfree() before anyway,
so if a flag is given multiple times memory leaks would have resulted.

OK joris@

Rename RCSFILE.fd to RCSFILE.rf_fd, plus minor spacing nits. No
binary change.

OK xsa@

Remove rcs_statfile(). It was just a wrapper for rcs_choosefile(),
handling exceptions oddly.

OK joris@

fix printf's

fix informative msg's wrt to -q and stderr. now matches gnu/usr.bin/rcs.

Remove /* NOTREACHED */ comments, now that lint is smart about
__dead. I never should have put them there.

OK xsa@.

o GNU ci returns 0 when a file is reverted (co -l file; ci file).
o Don't unlock file after revert.
o Fix spelling.

Passes extra tests in GNU rcstest.

OK niallo@.

Check return values for all strlcpy, and strlcat calls.

OK xsa@ and probably others.

prevent file races by obtaining an fd for the RCS file and
do our operations on that, this is safe and guarantees we
can operate on the file until we close(2) it.

a fix is coming for the remaining races in our diff code.
okay niallo@ and ray@

fork our code we shared between openrcs/cvs into the openrcs dir.

this was starting to become inhuman to maintain without
ugly ugly hacks in the shared code, and it will be easier
to make specific changes for openrcs without touching the
soon-to-be-replaced opencvs code.

fatal() -> err()/errx() as we try to not depend on heavy cvs_log().
OK joriski.

be consistent in error messages for wrongly specified keyword substitution
mode for `-k'.

move shared functions into rcsutil.[ch]; this makes rcsprog.c cleaner;
"the voices in my head say OK!" joris@.

remove overkill cvs_log() and rather use warn()/warnx().
OK ray@ and discussed with joris@.

switch -p handling to bitwise flags; OK ray@.

switch -q handling to bitwise flags; OK ray@ niallo@.

Plug some memory leaks; ok niallo@

excessive addition to brackets is a mental disease. you will go
blind, too. seek help. no binary change, ok joris

correctly handle RCS files without any revisions;
original diff from ray@, crafted into shape by myself.

okay ray@

*** empty log message ***

spaces

Clean up <rev> handling. Whenever a revision is specified after a
flag, it calls one of two new functions: rcs_setrevstr() or
rcs_setrevstr2(). rcs_setrevstr() sets a string to another string,
and complains if it was set more than once. rcs_setrevstr2() takes
two strings, sets one after the other, and fatal()s if more than
two strings were given.

All <rev> handling is now done in the loop that goes through each
argv. This is necessary for parsing symbols, which will be much
easier after this.

Along the way a lot of memory leaks were cleaned up. There is one
area where rcs_set_rev() is called, which allocates a RCSNUM and
stores it in pb.newrev, but it segfaults whenever I try to rcsnum_free()
it. I put an /* XXX */ comment there for now.

Passes regression tests and the code is less complicated in some
ways (to me).

Suggestions and OK xsa@

change rcs_kwexp_set() return type to void and simplify its use. OK niallo@.

- don't print out warnings about locking when they are actually wrong and
just confuse the user.

ok and input joris@

-z can take no argument.

first part of supporting branches in openrcs. right now we can only
check them out. commit is not working yet, but will be soon.

tested by myself and ray@
okay ray@

More closely matches GNU behavior, passes more GNU rcstest tests.

OK niallo@

- properly implement GNU file modes. basically, checkout will inherit
permissions from rcs file and initial checkin will inherit permissions from
working file.

problem spotted by uwe@

ok ray@ xsa@

Allow the -x flag to be used without an extension.

``that makes sense'' xsa@

- co usage() says we support -I, so set the flags and revision.

ok xsa@

- checkout_rev() can fail, in which case we want a non-zero exit code.

ok joris@

error message fix; respect -q too. Ok ray@.

Follow exit() with /* NOTREACHED*/ for lint.

``OK niallo@'' niallo@

Currently co(1) overwrites files that the current user doesn't have
write permissions for. If you own the directory, that means your
existing file, mode 000, is blown away without prompting you. GNU
RCS does this as well.

This fixes changes co(1) to always prompt if a file exists. This
breaks GNU compatibility but I think it's more important that we
prevent co(1) from overwriting files with strict permissions without
even prompting.

``makes sense'' xsa to an earlier diff

- use shared cvs_yesno() for yesno prompt, fixing an issue spotted by
deraadt@

- while here, remove unecessary cast also spotted by deraadt@

ok ray@ xsa@

-z support for RCS;

missing informative message after call to checkout_rev().

should also have removed those newlines in previous commit.

missing newline in checkout_rev(); spotted by joris@.

our deltatexts were sometimes different from gnu. rcs_getrev() was
unconditionally doing keyword expansion, even when this wasn't what we
wanted e.g. in checkout_rev() and cvs_checkout_rev(). so:

- change semantics of rcs_getrev() to not do keyword expansion
- add an rcs.c api function rcs_kwexp_buf() which does this
- change both checkout_rev() and cvs_checkout_rev() to use this function.

eyeballed by xsa, joris and Ray Lai

remove extra printf;

- fix behaviour when using -u and -l at the same time
- do not display some messages when checking in a file for the first time
OK niallo@.

cleanup, move log.h and rcs.h inclusion into rcsprog.h;

#include's cleanup; ok joris@ niallo@.

- implement lazy-parsing of rcs files, that is only parse as much as we
need. this can save us much work, particularly with very large rcs files.

first of a few important performance improvements.

ok joris@

support co -d, which checks out the first revision who's
date is less than or equal to the given date.

use fatal() if getlogin() fails;

switch to xmalloc stuff, me and xsa@ agreed on this a long
time ago, but we were being held back by jfb. too bad for him.

next step is to use fatal() through out the code for unrecoverable
errors instead of trying to be all nice and fluffy and reach main() again.

ok niallo@ and xsa@

if we are removing a lock that's not present, just check
out the file without complaining;

unused vars, my bad;

complete and correct rcs locking functionality,
it was only done partially and as a bonus, completely wrong.

seriously guys what was up with that?

correctly handle `-wfoo' and `-sbar' errors; OK joris@.

`-mmsg' does not belong here;

check for ownership of already present writable file when checking out;

writeable -> writable

consistency in variables naming;

some more bits for `-w[user]' support;

add support for `-T' and enable it for co(1);
tested and OK niallo@. "Looks good" joris@.

memset();

cannot combine -kv and -l;

initial bits for -T support;

add support for `-kmode';

sync SYNOPSIS with Reality and usage() accordingly;

make `-p' work again as expected;

enable `-xsuffixes' support as it has been added a few days ago
in rcs_statfile();
niallo ok

nuke trailing whitespaces;

- update comment for checkout_rev() for bitwise flags.

- sync checkout_usage();

- implement `-M' option.

- update the comment for checkout_state(); it did not tell the truth
about values returned.

changes from Venice:

- instead of using lots of individual flag variables, use a single int and
bitwise operations on it. while this saves memory, really it saves code
space and reduces complexity. checkout_rev() and checkout_state() in
particular benefit from these changes since their parameters where growing
really long.

- implement `-s' option in co

- implement `-M' option in ci

"I like this, go for it" joris@

nuke trailing whitespaces;

fix output if -p is specified; joris ok

- make a bunch of changes to how we handle verbose output. this brings us
much closer in line with GNU RCS and cleans things up in general.

ok joris@, "looks fine to me" xsa@

add co -p support;
input from xsa@ and niallo@;

use rcs_set_rev() instead of duplicating code all over the
place to set a variable to the given revision number;

support -f flag for co;
'fine' niallo@

spacing;

- sync checkout_usage();

- if revision passed to checkout_rev() is > HEAD, assume HEAD
(like GNU RCS).
- move verbose output into checkout_rev() function itself.

"looks sane" joris@

modularise checkout code into checkout_rev(). this shaves off a number
of lines in ci.c due to code re-use, and will simplify features which are in
the pipeline. no functional changes.

ok joris@

To be fully compatibly with the GNU RCS tools we need to have the
same way of parsing commandline options. Since getopt(3) allows spaces
between arguments and GNU RCS tools does not we needed to roll out our
own way of option handling, and here it is.

ok niallo@

minor knf;

various usage cleanup; ok joris

getlogin(2) sets errno;

sync checkout_usage();

- set permissions on checked out files correctly
- support bare `-u' and `-l' options which will check out HEAD.

note that with OpenRCS you must do `co -l -- <file>' and `co -u -- <file>'
when using bare arguments because our getopt(3) is POSIX compliant.

ok joris@

sync usage and add V flag;

- stop gcc's yapping;
- better output when locking and unlocking revisions;

- better output
- free frev when done

- support -l and -u flags;
- better output;

support -q flag in 'co' and 'rcs'

fix usage

put it basic checkout code

There was an extra newline in some cases and missing one in others.
OK joris@

Delay keyword expansion in co until after the locks have been resolved.

This way the expansion can happen with the correct values and files do not
show up as modified.

kill const and string artithmetic related warnings; ok millert@

Style fixes; from Ilya Kaliman

Convert xfree to free. From Fritjof Bornebusch. ok deraadt

Replace <sys/param.h> with <limits.h> and other less dirty headers where
possible. Annotate <sys/param.h> lines with their current reasons. Switch
to PATH_MAX, NGROUPS_MAX, HOST_NAME_MAX+1, LOGIN_NAME_MAX, etc. Change
MIN() and MAX() to local definitions of MINIMUM() and MAXIMUM() where
sensible to avoid pulling in the pollution. These are the files confirmed
through binary verification.
ok guenther, millert, doug (helped with the verification protocol)

Zap some stored values never read; From Fritjof Bornebusch; ok nicm@

exit in usage functions themselves; from Fritjof Bornebusch; ok nicm@

remove wrong casts; ok millert

add missing header needed by futimes()

ok stsp@

Remove the need for rp_file in parser structure, instead keep only one
FILE pointer in RCSFILE. This fixes some ugliness in closing an
fdopen()ed FILE and its underlying file descriptor.

Notified by Joerg Sonnenberger <joerg at britannica dot bec to de>

discussed with and ok nicm

Set rcs_suffixes to default value on initialization.

ok ray zinovik

Check date_parse return values.

OK xsa

Make BUFs autoextend by default. This was already done in OpenCVS
a while ago.

OK zinovik nicm

Reduce variable/function name and whitespace differences between
cvs/rcs.

OK xsa zinovik

Checking if a file is ASCII should only be done when diffing, because
the non-ASCII characters could be printed to the screen. For
checking in files, checking out files, merging files, and removing
revisions of files, we do so regardless of whether the files are
ASCII or binary.

Fixes PR6031.

OK joris and deraadt.

Rename rcs_diffreg() to diffreg().

OK joris@

Sync revisions and time buffers size to be consistent with each others.
Simplifies further size tweaks if needed.
OK ray@.

If strict locking is disabled, make file writable by owner on checkout.
Matches GNU's behaviour. From Pierre Riteau (long time ago).

Correct exit value when one of the files specified cannot be checked out.
Matches GNU's behaviour. From Pierre Riteau (long time ago).

As done in OpenCVS, general includes cleanup sweep. OK otto@.

If a ,suffix file is given as an arg to ci and co, strip it. Avoids
potential disasters. Initial diff from niallo@, ok niallo@ joris@

make sure checkout_rev() sets correct file permissions via file->rf_mode
in case where fstat() of file->rf_fd won't give us right mode.

fixes PR #5383

ok xsa@

keep correct permissions on the RCS file when locking/unlocking the file
with ci and co. fixes PR 5379.
OK millert@ joris@.

When checking to see if a checked-out copy of the file is the same
as what is in the repo, expand keywords before doing the diff.
OK xsa@

Unlike GNU rcs, our co will not overwrite an existing file, even
if the existing file is not writable. While safer, this can be
annoying. I've changed things to compare the version of the file
being checked out with its current contents and, if they are the
same, to allow the checkout without user intervention. The behavior
when there is a writable version of the file is unchanged. OK xsa@

Simplify stripping of write bits from file mode.
Add support for reusing the checkin message for multiple files, ala GNU
Fix the message when you abort a checkout and the file was not writable.
OK joris@ niallo@

- convert some warnx() to warn(), which fixes a few stupidly un-informative error messages (found by tom@)
- make rcs_choosefile() save wrt errno, input from ray@

ok joris@ xsa@

More strlcpy/strlcat -> snprintf/xasprintf cleanup.

OK joris@

Reduce rcs_buf_release() usage, improving binary support.

OK niallo@

- correctly handle binary files; say bye bye to using c strings for deltatexts.

"slap it in" joris@

Remove a lot of xstrdup() calls in getopt() loops, which are usually
unnecessary. These xstrdup() calls don't call xfree() before anyway,
so if a flag is given multiple times memory leaks would have resulted.

OK joris@

Rename RCSFILE.fd to RCSFILE.rf_fd, plus minor spacing nits. No
binary change.

OK xsa@

Remove rcs_statfile(). It was just a wrapper for rcs_choosefile(),
handling exceptions oddly.

OK joris@

fix printf's

fix informative msg's wrt to -q and stderr. now matches gnu/usr.bin/rcs.

Remove /* NOTREACHED */ comments, now that lint is smart about
__dead. I never should have put them there.

OK xsa@.

o GNU ci returns 0 when a file is reverted (co -l file; ci file).
o Don't unlock file after revert.
o Fix spelling.

Passes extra tests in GNU rcstest.

OK niallo@.

Check return values for all strlcpy, and strlcat calls.

OK xsa@ and probably others.

prevent file races by obtaining an fd for the RCS file and
do our operations on that, this is safe and guarantees we
can operate on the file until we close(2) it.

a fix is coming for the remaining races in our diff code.
okay niallo@ and ray@

fork our code we shared between openrcs/cvs into the openrcs dir.

this was starting to become inhuman to maintain without
ugly ugly hacks in the shared code, and it will be easier
to make specific changes for openrcs without touching the
soon-to-be-replaced opencvs code.

fatal() -> err()/errx() as we try to not depend on heavy cvs_log().
OK joriski.

be consistent in error messages for wrongly specified keyword substitution
mode for `-k'.

move shared functions into rcsutil.[ch]; this makes rcsprog.c cleaner;
"the voices in my head say OK!" joris@.

remove overkill cvs_log() and rather use warn()/warnx().
OK ray@ and discussed with joris@.

switch -p handling to bitwise flags; OK ray@.

switch -q handling to bitwise flags; OK ray@ niallo@.

Plug some memory leaks; ok niallo@

excessive addition to brackets is a mental disease. you will go
blind, too. seek help. no binary change, ok joris

correctly handle RCS files without any revisions;
original diff from ray@, crafted into shape by myself.

okay ray@

*** empty log message ***

spaces

Clean up <rev> handling. Whenever a revision is specified after a
flag, it calls one of two new functions: rcs_setrevstr() or
rcs_setrevstr2(). rcs_setrevstr() sets a string to another string,
and complains if it was set more than once. rcs_setrevstr2() takes
two strings, sets one after the other, and fatal()s if more than
two strings were given.

All <rev> handling is now done in the loop that goes through each
argv. This is necessary for parsing symbols, which will be much
easier after this.

Along the way a lot of memory leaks were cleaned up. There is one
area where rcs_set_rev() is called, which allocates a RCSNUM and
stores it in pb.newrev, but it segfaults whenever I try to rcsnum_free()
it. I put an /* XXX */ comment there for now.

Passes regression tests and the code is less complicated in some
ways (to me).

Suggestions and OK xsa@

change rcs_kwexp_set() return type to void and simplify its use. OK niallo@.

- don't print out warnings about locking when they are actually wrong and
just confuse the user.

ok and input joris@

-z can take no argument.

first part of supporting branches in openrcs. right now we can only
check them out. commit is not working yet, but will be soon.

tested by myself and ray@
okay ray@

More closely matches GNU behavior, passes more GNU rcstest tests.

OK niallo@

- properly implement GNU file modes. basically, checkout will inherit
permissions from rcs file and initial checkin will inherit permissions from
working file.

problem spotted by uwe@

ok ray@ xsa@

Allow the -x flag to be used without an extension.

``that makes sense'' xsa@

- co usage() says we support -I, so set the flags and revision.

ok xsa@

- checkout_rev() can fail, in which case we want a non-zero exit code.

ok joris@

error message fix; respect -q too. Ok ray@.

Follow exit() with /* NOTREACHED*/ for lint.

``OK niallo@'' niallo@

Currently co(1) overwrites files that the current user doesn't have
write permissions for. If you own the directory, that means your
existing file, mode 000, is blown away without prompting you. GNU
RCS does this as well.

This fixes changes co(1) to always prompt if a file exists. This
breaks GNU compatibility but I think it's more important that we
prevent co(1) from overwriting files with strict permissions without
even prompting.

``makes sense'' xsa to an earlier diff

- use shared cvs_yesno() for yesno prompt, fixing an issue spotted by
deraadt@

- while here, remove unecessary cast also spotted by deraadt@

ok ray@ xsa@

-z support for RCS;

missing informative message after call to checkout_rev().

should also have removed those newlines in previous commit.

missing newline in checkout_rev(); spotted by joris@.

our deltatexts were sometimes different from gnu. rcs_getrev() was
unconditionally doing keyword expansion, even when this wasn't what we
wanted e.g. in checkout_rev() and cvs_checkout_rev(). so:

- change semantics of rcs_getrev() to not do keyword expansion
- add an rcs.c api function rcs_kwexp_buf() which does this
- change both checkout_rev() and cvs_checkout_rev() to use this function.

eyeballed by xsa, joris and Ray Lai

remove extra printf;

- fix behaviour when using -u and -l at the same time
- do not display some messages when checking in a file for the first time
OK niallo@.

cleanup, move log.h and rcs.h inclusion into rcsprog.h;

#include's cleanup; ok joris@ niallo@.

- implement lazy-parsing of rcs files, that is only parse as much as we
need. this can save us much work, particularly with very large rcs files.

first of a few important performance improvements.

ok joris@

support co -d, which checks out the first revision who's
date is less than or equal to the given date.

use fatal() if getlogin() fails;

switch to xmalloc stuff, me and xsa@ agreed on this a long
time ago, but we were being held back by jfb. too bad for him.

next step is to use fatal() through out the code for unrecoverable
errors instead of trying to be all nice and fluffy and reach main() again.

ok niallo@ and xsa@

if we are removing a lock that's not present, just check
out the file without complaining;

unused vars, my bad;

complete and correct rcs locking functionality,
it was only done partially and as a bonus, completely wrong.

seriously guys what was up with that?

correctly handle `-wfoo' and `-sbar' errors; OK joris@.

`-mmsg' does not belong here;

check for ownership of already present writable file when checking out;

writeable -> writable

consistency in variables naming;

some more bits for `-w[user]' support;

add support for `-T' and enable it for co(1);
tested and OK niallo@. "Looks good" joris@.

memset();

cannot combine -kv and -l;

initial bits for -T support;

add support for `-kmode';

sync SYNOPSIS with Reality and usage() accordingly;

make `-p' work again as expected;

enable `-xsuffixes' support as it has been added a few days ago
in rcs_statfile();
niallo ok

nuke trailing whitespaces;

- update comment for checkout_rev() for bitwise flags.

- sync checkout_usage();

- implement `-M' option.

- update the comment for checkout_state(); it did not tell the truth
about values returned.

changes from Venice:

- instead of using lots of individual flag variables, use a single int and
bitwise operations on it. while this saves memory, really it saves code
space and reduces complexity. checkout_rev() and checkout_state() in
particular benefit from these changes since their parameters where growing
really long.

- implement `-s' option in co

- implement `-M' option in ci

"I like this, go for it" joris@

nuke trailing whitespaces;

fix output if -p is specified; joris ok

- make a bunch of changes to how we handle verbose output. this brings us
much closer in line with GNU RCS and cleans things up in general.

ok joris@, "looks fine to me" xsa@

add co -p support;
input from xsa@ and niallo@;

use rcs_set_rev() instead of duplicating code all over the
place to set a variable to the given revision number;

support -f flag for co;
'fine' niallo@

spacing;

- sync checkout_usage();

- if revision passed to checkout_rev() is > HEAD, assume HEAD
(like GNU RCS).
- move verbose output into checkout_rev() function itself.

"looks sane" joris@

modularise checkout code into checkout_rev(). this shaves off a number
of lines in ci.c due to code re-use, and will simplify features which are in
the pipeline. no functional changes.

ok joris@

To be fully compatibly with the GNU RCS tools we need to have the
same way of parsing commandline options. Since getopt(3) allows spaces
between arguments and GNU RCS tools does not we needed to roll out our
own way of option handling, and here it is.

ok niallo@

minor knf;

various usage cleanup; ok joris

getlogin(2) sets errno;

sync checkout_usage();

- set permissions on checked out files correctly
- support bare `-u' and `-l' options which will check out HEAD.

note that with OpenRCS you must do `co -l -- <file>' and `co -u -- <file>'
when using bare arguments because our getopt(3) is POSIX compliant.

ok joris@

sync usage and add V flag;

- stop gcc's yapping;
- better output when locking and unlocking revisions;

- better output
- free frev when done

- support -l and -u flags;
- better output;

support -q flag in 'co' and 'rcs'

fix usage

put it basic checkout code

Delay keyword expansion in co until after the locks have been resolved.

This way the expansion can happen with the correct values and files do not
show up as modified.

kill const and string artithmetic related warnings; ok millert@

Style fixes; from Ilya Kaliman

Convert xfree to free. From Fritjof Bornebusch. ok deraadt

Replace <sys/param.h> with <limits.h> and other less dirty headers where
possible. Annotate <sys/param.h> lines with their current reasons. Switch
to PATH_MAX, NGROUPS_MAX, HOST_NAME_MAX+1, LOGIN_NAME_MAX, etc. Change
MIN() and MAX() to local definitions of MINIMUM() and MAXIMUM() where
sensible to avoid pulling in the pollution. These are the files confirmed
through binary verification.
ok guenther, millert, doug (helped with the verification protocol)

Zap some stored values never read; From Fritjof Bornebusch; ok nicm@

exit in usage functions themselves; from Fritjof Bornebusch; ok nicm@

remove wrong casts; ok millert

add missing header needed by futimes()

ok stsp@

Remove the need for rp_file in parser structure, instead keep only one
FILE pointer in RCSFILE. This fixes some ugliness in closing an
fdopen()ed FILE and its underlying file descriptor.

Notified by Joerg Sonnenberger <joerg at britannica dot bec to de>

discussed with and ok nicm

Set rcs_suffixes to default value on initialization.

ok ray zinovik

Check date_parse return values.

OK xsa

Make BUFs autoextend by default. This was already done in OpenCVS
a while ago.

OK zinovik nicm

Reduce variable/function name and whitespace differences between
cvs/rcs.

OK xsa zinovik

Checking if a file is ASCII should only be done when diffing, because
the non-ASCII characters could be printed to the screen. For
checking in files, checking out files, merging files, and removing
revisions of files, we do so regardless of whether the files are
ASCII or binary.

Fixes PR6031.

OK joris and deraadt.

Rename rcs_diffreg() to diffreg().

OK joris@

Sync revisions and time buffers size to be consistent with each others.
Simplifies further size tweaks if needed.
OK ray@.

If strict locking is disabled, make file writable by owner on checkout.
Matches GNU's behaviour. From Pierre Riteau (long time ago).

Correct exit value when one of the files specified cannot be checked out.
Matches GNU's behaviour. From Pierre Riteau (long time ago).

As done in OpenCVS, general includes cleanup sweep. OK otto@.

If a ,suffix file is given as an arg to ci and co, strip it. Avoids
potential disasters. Initial diff from niallo@, ok niallo@ joris@

make sure checkout_rev() sets correct file permissions via file->rf_mode
in case where fstat() of file->rf_fd won't give us right mode.

fixes PR #5383

ok xsa@

keep correct permissions on the RCS file when locking/unlocking the file
with ci and co. fixes PR 5379.
OK millert@ joris@.

When checking to see if a checked-out copy of the file is the same
as what is in the repo, expand keywords before doing the diff.
OK xsa@

Unlike GNU rcs, our co will not overwrite an existing file, even
if the existing file is not writable. While safer, this can be
annoying. I've changed things to compare the version of the file
being checked out with its current contents and, if they are the
same, to allow the checkout without user intervention. The behavior
when there is a writable version of the file is unchanged. OK xsa@

Simplify stripping of write bits from file mode.
Add support for reusing the checkin message for multiple files, ala GNU
Fix the message when you abort a checkout and the file was not writable.
OK joris@ niallo@

- convert some warnx() to warn(), which fixes a few stupidly un-informative error messages (found by tom@)
- make rcs_choosefile() save wrt errno, input from ray@

ok joris@ xsa@

More strlcpy/strlcat -> snprintf/xasprintf cleanup.

OK joris@

Reduce rcs_buf_release() usage, improving binary support.

OK niallo@

- correctly handle binary files; say bye bye to using c strings for deltatexts.

"slap it in" joris@

Remove a lot of xstrdup() calls in getopt() loops, which are usually
unnecessary. These xstrdup() calls don't call xfree() before anyway,
so if a flag is given multiple times memory leaks would have resulted.

OK joris@

Rename RCSFILE.fd to RCSFILE.rf_fd, plus minor spacing nits. No
binary change.

OK xsa@

Remove rcs_statfile(). It was just a wrapper for rcs_choosefile(),
handling exceptions oddly.

OK joris@

fix printf's

fix informative msg's wrt to -q and stderr. now matches gnu/usr.bin/rcs.

Remove /* NOTREACHED */ comments, now that lint is smart about
__dead. I never should have put them there.

OK xsa@.

o GNU ci returns 0 when a file is reverted (co -l file; ci file).
o Don't unlock file after revert.
o Fix spelling.

Passes extra tests in GNU rcstest.

OK niallo@.

Check return values for all strlcpy, and strlcat calls.

OK xsa@ and probably others.

prevent file races by obtaining an fd for the RCS file and
do our operations on that, this is safe and guarantees we
can operate on the file until we close(2) it.

a fix is coming for the remaining races in our diff code.
okay niallo@ and ray@

fork our code we shared between openrcs/cvs into the openrcs dir.

this was starting to become inhuman to maintain without
ugly ugly hacks in the shared code, and it will be easier
to make specific changes for openrcs without touching the
soon-to-be-replaced opencvs code.

fatal() -> err()/errx() as we try to not depend on heavy cvs_log().
OK joriski.

be consistent in error messages for wrongly specified keyword substitution
mode for `-k'.

move shared functions into rcsutil.[ch]; this makes rcsprog.c cleaner;
"the voices in my head say OK!" joris@.

remove overkill cvs_log() and rather use warn()/warnx().
OK ray@ and discussed with joris@.

switch -p handling to bitwise flags; OK ray@.

switch -q handling to bitwise flags; OK ray@ niallo@.

Plug some memory leaks; ok niallo@

excessive addition to brackets is a mental disease. you will go
blind, too. seek help. no binary change, ok joris

correctly handle RCS files without any revisions;
original diff from ray@, crafted into shape by myself.

okay ray@

*** empty log message ***

spaces

Clean up <rev> handling. Whenever a revision is specified after a
flag, it calls one of two new functions: rcs_setrevstr() or
rcs_setrevstr2(). rcs_setrevstr() sets a string to another string,
and complains if it was set more than once. rcs_setrevstr2() takes
two strings, sets one after the other, and fatal()s if more than
two strings were given.

All <rev> handling is now done in the loop that goes through each
argv. This is necessary for parsing symbols, which will be much
easier after this.

Along the way a lot of memory leaks were cleaned up. There is one
area where rcs_set_rev() is called, which allocates a RCSNUM and
stores it in pb.newrev, but it segfaults whenever I try to rcsnum_free()
it. I put an /* XXX */ comment there for now.

Passes regression tests and the code is less complicated in some
ways (to me).

Suggestions and OK xsa@

change rcs_kwexp_set() return type to void and simplify its use. OK niallo@.

- don't print out warnings about locking when they are actually wrong and
just confuse the user.

ok and input joris@

-z can take no argument.

first part of supporting branches in openrcs. right now we can only
check them out. commit is not working yet, but will be soon.

tested by myself and ray@
okay ray@

More closely matches GNU behavior, passes more GNU rcstest tests.

OK niallo@

- properly implement GNU file modes. basically, checkout will inherit
permissions from rcs file and initial checkin will inherit permissions from
working file.

problem spotted by uwe@

ok ray@ xsa@

Allow the -x flag to be used without an extension.

``that makes sense'' xsa@

- co usage() says we support -I, so set the flags and revision.

ok xsa@

- checkout_rev() can fail, in which case we want a non-zero exit code.

ok joris@

error message fix; respect -q too. Ok ray@.

Follow exit() with /* NOTREACHED*/ for lint.

``OK niallo@'' niallo@

Currently co(1) overwrites files that the current user doesn't have
write permissions for. If you own the directory, that means your
existing file, mode 000, is blown away without prompting you. GNU
RCS does this as well.

This fixes changes co(1) to always prompt if a file exists. This
breaks GNU compatibility but I think it's more important that we
prevent co(1) from overwriting files with strict permissions without
even prompting.

``makes sense'' xsa to an earlier diff

- use shared cvs_yesno() for yesno prompt, fixing an issue spotted by
deraadt@

- while here, remove unecessary cast also spotted by deraadt@

ok ray@ xsa@

-z support for RCS;

missing informative message after call to checkout_rev().

should also have removed those newlines in previous commit.

missing newline in checkout_rev(); spotted by joris@.

our deltatexts were sometimes different from gnu. rcs_getrev() was
unconditionally doing keyword expansion, even when this wasn't what we
wanted e.g. in checkout_rev() and cvs_checkout_rev(). so:

- change semantics of rcs_getrev() to not do keyword expansion
- add an rcs.c api function rcs_kwexp_buf() which does this
- change both checkout_rev() and cvs_checkout_rev() to use this function.

eyeballed by xsa, joris and Ray Lai

remove extra printf;

- fix behaviour when using -u and -l at the same time
- do not display some messages when checking in a file for the first time
OK niallo@.

cleanup, move log.h and rcs.h inclusion into rcsprog.h;

#include's cleanup; ok joris@ niallo@.

- implement lazy-parsing of rcs files, that is only parse as much as we
need. this can save us much work, particularly with very large rcs files.

first of a few important performance improvements.

ok joris@

support co -d, which checks out the first revision who's
date is less than or equal to the given date.

use fatal() if getlogin() fails;

switch to xmalloc stuff, me and xsa@ agreed on this a long
time ago, but we were being held back by jfb. too bad for him.

next step is to use fatal() through out the code for unrecoverable
errors instead of trying to be all nice and fluffy and reach main() again.

ok niallo@ and xsa@