Replace use of the old BSD st_*timespec members in struct stat with
the POSIX-standard st_*tim members.

ok millert@

don't access(conffile)

This removes a few access(2) calls in the configuration file handling.
startupfile() now opens and return the file and to avoid surprises it
also uses a caller-provided buffer to store the filename. This also
removes the extra adjustpath() that load() did: it has been moved to
evalfile() only.

with help, fixes and ok tb@

strncpy -> strlcpy to properly NUL-terminate the copy of the path

otherwise paths longer than NFILEN (1024) given with -u won't
NUL-terminate `file'.

Delete obsolete /* ARGSUSED */ lint comments.

ok miod@ millert@

Put the hardcoded '\n' character which is found throughout mg into a
buffer specific variable. The diff should not produce any behavourial
changes in mg.

Allow the user to specify a path to the mg startup file on the command line.

This diff moves the check for a newline at the end of a buffer (when
saving it) earlier. This means data doesn't 'disappear' from the file
on disk while the user answers 'yes' or 'no'.

mg tries to write backups to the wrong directory when run under a
different effective user, i.e. when invoced via su and
backup-to-home-directory is enabled.

Problem pointed out and diff provied by Lucas Gabriel Vuotto
<lvuotto92 () gmail ! com>, thanks!

Subsequently slacked on for nearly a year by yours truly.

Then remembered when Han Boetes <hboetes () gmail ! com> came up with
a similar diff because of a problem report by Mark Willson where it
turned out that getlogin(2) is not very portable.

OK tb

fix -Wshadow warnings.
Found by bcallah, thanks!

strncmp is a more reasonable way to check the beginning of a string.

these programs probably do not need to use TMPDIR. ok florian

Use fstatat() instead of crafting a filename to use with stat()

ok millert@

ensure the backup file has the same mtime as the original file, this is in line
with emacs' behaviour.

from Harald Dunkel via Han Boetes
ok lum@

Clean up the includes in mg.
This does the following:
Moves all POSIX headers from sysdef.h into the individual .c files so that
each file now only includes what it needs. All headers are properly sorted.
Moves the remainder of sysdef.h to other files (mostly def.h) and deletes
sysdef.h now that it's no longer contains anything.
Tweak a comment that references sysdef.h so that it no longer does that.
ok florian@

Stop using <sys/param.h>; replace MAXPATHLEN with PATH_MAX, stop using MAX(),
and pull in <limits.h> for *_MAX constants.

inspired on a diff from Kamil Rytarowski (n54 (at) gmx.com)
ok bcallah@

Add some missing dobeeps.
ok florian@

and and -> and

- move prototype of expandtilder() to def.h; needed by an upcoming diff.

Fix an outstanding debian bug: #658539

"looks good" kjell

- add a pathnames.h
- some whitespace cleanup

ok lum@

Fixes a problem where if you try to open a file that doesn't exist and has
a name longer than LOGIN_NAME_MAX and also has a tilde at the front
e.g:

$ mg ~01234567890123456789012345678901

mg will give a "Login name too long" instead of opening a new buffer
named ~01234567890123456789012345678901

Allow mg to save backup files to a users home directory.
Suggestions from eric@ and Sunil Nimmagadda.
Remarks deraadt@.

Change error message to reflect the correct filename: nname is 'New
name' as opposed to tname 'Temp name'

Remove static FILE pointer used for handling files in fileio.c. Pass
by reference instead. This allows the mg startup file to open other
files without unexpected things happening.

Discussed with Sunil Nimmagadda.

Change how mg handles files beginning with a '~'.
Bring behaviour more into line with emacs.

Remove the conditional directives NO_MACRO and NO_STARTUP.
They have not compiled for numerous years.
ok kjell@ millert@

Close an opendir(). From Igor Zinovik.

On a file write fail:

1. return an error value
2. show an error message

From Loganaden Velvindron with suggestion from millert@

It volates style(9), but in mg, #include"def.h" goes first. 'twas the way it was built. no binary change here. confirmed by lum@, tested by Henri Kemppainen

Change len+memcmp check to a strncmp for prefix testing on
filename completion. Avoids a recurring portability headache.
ok lum@
"concerns assuaged to the soothing sounds of Martin Denny blambert@"
(I worry about that guy...)

Enable dirty buffer detection in mg.
Emulate the emacs behavior: after suspend/resume, buffer switch,
or at save time, warn (prompt) the user if the file has been modified
on disk in the interim.
This has already saved my butt numerous times.
ok phessler

typos; ok jmc@
sys/dev/pci/pciide.c from naddy@

Fix a bug where inserting a file resulted in an incorrect
line-number count for a buffer (M-X insert-file, M-> to reproduce).
While here, fix a number of bugs with incorrect line numbers
after swap point-and-mark

Originally reported via debian's bug tracking system. Fix tested by
Han Boetes and Deanna Phillips.

spacing

Use S_IS* macros insted of masking with S_IF* flags. The latter may
have multiple bits set, which lead to surprising results. Spotted by
Paul Stoeber, more to come. ok millert@ pedro@ jaredy@ djm@

Rename the header line of a buffer to b_headp, from the remarkably
unintuitive b_linep. No binary change.

- use <dirent.h> not <sys/dir.h> (this appeases FreeBSD's annoying #warning)
- grab <sys/time.h> before <sys/resource.h> like getrusage(2) says
ok kjell

fix style

make // /~ path rewriting optional in adjustname() and use it everywhere
except for the command line specified files. ok kjell,cloder

correct one more comment

Do some KNF, clean up some unused junk that has lying around for 40-odd
cvs revisions, and fix a few comments to match reality

len = strlen(foo); if foo[len - 1]... will do bad things if len == 0.
Avoid, and rearrange a test so it can't (hypothetically) overflow.

lint love; ok kjell

test char against '\0', not 0

Clean up the ugly casted frees. In one case, this meant eliminating a nasty
struct/union/casting nightmare when building the list of names for filename
completion. In particular, be consistent about strduping and freeing
the list data.

Do some delinting of strl-type functions. Also, remove a superfluous
word in the undo-list.

do some silly de-linting

#ifdef NO_DIR can go too. From Han Boetes.

Kill the NO_BACKUP #ifdef

toast NO_DIRED #ifdef; ok kjell

greedy use of typedef struct was making code harder to read; ok kjell cloder

Better error checking of snprintfs. From Han Boetes.

Now that we have an editable minibuffer, revert parsing of
"//" and "/~" as '/' and '~' respectively. I know emacs does it,
but it is weird, and breaks things that foolishly use filenames
like "/tmp//crontab.xxxx".
Proddings from matthieu and deraadt. Error report from Bernd Ahlers.

spacing

When attempting to open a filename, walk backwards through minibuffer:
a> if you run into the beginning of the string, use the whole thing
b> if you run into a // combo, use everything starting from the second /
c> if you run into a /~ combo, use everything starting from the ~
i.e. do like emacs.
From (and for) Jason Wright

Fix string handling. LOGIN_NAME_MAX includes the NUL, so don't reserve extra
space. Fix usage of strlcpy. Correct test for terminating slash.

"looks good" deraadt@

more headers to please lint

move the dired routines into dired.c where they belong.
ok cloder@

Make dired buffer read-only by default.
Noticed by Han Boetes

Use dired mode automatically if file specified for loading is
a directory. Modified version of patch from Han Boetes.
ok cloder@

check if current line is null. fixes a core in dired.

Fix dired mode. Make deletions work, and keystrokes match emacs.
* dired-other-window should default to current buffer's path.
* Remove redundant code (from Han Boetes)
* Move initialization (and since we're here, fix mail-mode initialization
too)
* Remove redundant keymap in dired.c, and replace it with
the better one that was being ignored in keymap.c.

Add explicit public domain notices to all public domain files.
ok millert@, deraadt@

This is a no binary change which does:

- spelling, punctuation fixes
- variable declaration lineup
- use parentheses for return and sizeof
- K&R function declarations -> ANSI
- other minor code beautification

ok henning@

spacing

the realpath() was needed to do do relative to absolute path
conversion. revert my diff that takes it out, and instead, on
realpath() failure (like when you have no perms on pwd) just return the
unexpanded path. everything still fine with that, just tab completeion
and the like does not work (obviously).
this was mainly for "sudo mg /etc/something" from ~ where ~ is
nfs-mounted with root mapped to -2
and now, that case works as well as tab completion on insert-file etc.

fix mg's behaviour with regards to files on which we do not have
write access. diff originally from vincent@ and forgotten for
a while.

ok rohee@, "toss it in, i'll bitch if it doesn't work" henning@

make fbackupfile() return FALSE when the rename(2) from the
mkstemp()-derived name to the filename~ one fails
From: Han Boetes <han@mijncomputer.nl>

Safer backup file generation using mkstemp(); henning@ OK

no need to call realpath() in adjustname()
realpath returns an error if cwd is not writeable, and that is quite common
with NFS-mounted ~ and sudo mg /etc/something - annoying as hell. realpath
is not needed at all here, so zap it. vincent ok

when saving a file, check whether it is newline terminated and if it is not,
prompt the user and add the newline if he agrees.

tested by a lot of helpful tech@ guys, approved by henning

simplify the file writing loop

looked at by ian, tested on i386, macppc, sparc64

read(2) returns ssize_t, not size_t. spotted by Andrey Matveev.

typos from Jonathon Gray;

make dired work more like emacs. (well, make it _work_, and then add a
bunch of missing commands)

actually extract the filename off the ls line. makes dired *slightly*
more useful.

use NULL to indicate failure in functions that return pointers, not FALSE.
typo in comment

from Patrick Latifi (patrick.l@hermes.usherb.ca)

fix a small memleak on errors. noticed by pat latifi.

ok millert, fgsch

knf

replace the ugly and buggy adjustname function by a simple one using
simple APIs. makes mg not crash with 65k filenames...

ok art@

KNF

fix bad usage of strlcpy()'s return value. (made pointers point
beyond the boundaries of buffers)

ok deraadt@

don't use /bin/cp to copy files.

ok deraadt@

fix mem leak

ok deraadt@

use fchmod and fchown instead; vincent

Fixes the the following problems (from zyrnix)
1) Mg crashes with column width of 1, rows > 2
2) Mg panics on resizing due to incomplete write
3) Mg doesn't compile with STARTUPFILE defined
deraadt@ OK

Make these special buffers readonly by default. ok art@

* Move to ANSI function definitions.
* Add a whole lot of consts where I thought it made sense

no ok, but no objections either...

if file mode known, try to avoid a race.. i think. millert spotted a bug in my first draft

d_makename now takes a length parameter so we can remove the
strcpy call.

some KNF

* Replace unsafe strcpy and strcat calls to safe strlcpy and strlcat.
* Be a little bit more verbose about some errors
* Fix some memory leaks in fileio.c

ok deraadt@, art@

correct fork logic and rewrite for clarity, ok todd@

correct type on last arg to execl(); nordin@cse.ogi.edu

Change the filename completion code to:
- not use an unprotected strcat.
- Append a '/' when a file is a symlink to a directory.
(tab completion on '/sys' is now less annoying.)
- actually make some more sense.

Don't assume that map_table[0] is the fundamental map.

spaces

use strlcpy vs cpy+len, some other fixes

Get rid of unnecessary casts of NULL.

remove VOID. We're way beyond the point where this could build with an
ancient compiler.

* Let make_file_list always append a '/' at the end of directories.
This makes filename TAB-completion more useful, now you don't need to
write that '/' manually.
* random cleanups including using snprintf and memcmp where open-coded
equivalents were used, waitpid instead of 'while (wait() != pid);', etc.

Use opendir/readdir instead of open/read when making a file list.
This makes filename tab-completion work and takes dired a step further.

$OpenBSD$

Don't use rename to backup the file to a ~ file. We need to make a
copy instead so the original is edited in place for things like
cron and vipw.

$HOME paranoia: never use getenv("HOME") w/o checking for NULL and non-zero

The start of KNF + -Wall. The code has been run through indent but
needs hand fixup. I stopped at keymap.c...

Remove a now-bogus comment
Use pid_t
Use vfork(), not fork() when sensible

We don't have silly SYSV 14-character file name limits. Taken from
the BSD fileio.c. Note that this module is rife with PATH_MAX overflow
possibilities.

Use uid_t, gid_t and mode_t where applicable.
Pull in changes from the BSD-specific sysdef.h and fileio.c:
kill bogus unlinkdir() and rename() kludges since we have rmdir(2) and rename(2)
define SYMBLINK since we have symlinks
TODO: merge BSD spawn.c and ttyio.c

further pruning

make it minimally compile; millert

initial import of mg2a

don't access(conffile)

This removes a few access(2) calls in the configuration file handling.
startupfile() now opens and return the file and to avoid surprises it
also uses a caller-provided buffer to store the filename. This also
removes the extra adjustpath() that load() did: it has been moved to
evalfile() only.

with help, fixes and ok tb@

strncpy -> strlcpy to properly NUL-terminate the copy of the path

otherwise paths longer than NFILEN (1024) given with -u won't
NUL-terminate `file'.

Delete obsolete /* ARGSUSED */ lint comments.

ok miod@ millert@

Put the hardcoded '\n' character which is found throughout mg into a
buffer specific variable. The diff should not produce any behavourial
changes in mg.

Allow the user to specify a path to the mg startup file on the command line.

This diff moves the check for a newline at the end of a buffer (when
saving it) earlier. This means data doesn't 'disappear' from the file
on disk while the user answers 'yes' or 'no'.

mg tries to write backups to the wrong directory when run under a
different effective user, i.e. when invoced via su and
backup-to-home-directory is enabled.

Problem pointed out and diff provied by Lucas Gabriel Vuotto
<lvuotto92 () gmail ! com>, thanks!

Subsequently slacked on for nearly a year by yours truly.

Then remembered when Han Boetes <hboetes () gmail ! com> came up with
a similar diff because of a problem report by Mark Willson where it
turned out that getlogin(2) is not very portable.

OK tb

fix -Wshadow warnings.
Found by bcallah, thanks!

strncmp is a more reasonable way to check the beginning of a string.

these programs probably do not need to use TMPDIR. ok florian

Use fstatat() instead of crafting a filename to use with stat()

ok millert@

ensure the backup file has the same mtime as the original file, this is in line
with emacs' behaviour.

from Harald Dunkel via Han Boetes
ok lum@

Clean up the includes in mg.
This does the following:
Moves all POSIX headers from sysdef.h into the individual .c files so that
each file now only includes what it needs. All headers are properly sorted.
Moves the remainder of sysdef.h to other files (mostly def.h) and deletes
sysdef.h now that it's no longer contains anything.
Tweak a comment that references sysdef.h so that it no longer does that.
ok florian@

Stop using <sys/param.h>; replace MAXPATHLEN with PATH_MAX, stop using MAX(),
and pull in <limits.h> for *_MAX constants.

inspired on a diff from Kamil Rytarowski (n54 (at) gmx.com)
ok bcallah@

Add some missing dobeeps.
ok florian@

and and -> and

- move prototype of expandtilder() to def.h; needed by an upcoming diff.

Fix an outstanding debian bug: #658539

"looks good" kjell

- add a pathnames.h
- some whitespace cleanup

ok lum@

Fixes a problem where if you try to open a file that doesn't exist and has
a name longer than LOGIN_NAME_MAX and also has a tilde at the front
e.g:

$ mg ~01234567890123456789012345678901

mg will give a "Login name too long" instead of opening a new buffer
named ~01234567890123456789012345678901

Allow mg to save backup files to a users home directory.
Suggestions from eric@ and Sunil Nimmagadda.
Remarks deraadt@.

Change error message to reflect the correct filename: nname is 'New
name' as opposed to tname 'Temp name'

Remove static FILE pointer used for handling files in fileio.c. Pass
by reference instead. This allows the mg startup file to open other
files without unexpected things happening.

Discussed with Sunil Nimmagadda.

Change how mg handles files beginning with a '~'.
Bring behaviour more into line with emacs.

Remove the conditional directives NO_MACRO and NO_STARTUP.
They have not compiled for numerous years.
ok kjell@ millert@

Close an opendir(). From Igor Zinovik.

On a file write fail:

1. return an error value
2. show an error message

From Loganaden Velvindron with suggestion from millert@

It volates style(9), but in mg, #include"def.h" goes first. 'twas the way it was built. no binary change here. confirmed by lum@, tested by Henri Kemppainen

Change len+memcmp check to a strncmp for prefix testing on
filename completion. Avoids a recurring portability headache.
ok lum@
"concerns assuaged to the soothing sounds of Martin Denny blambert@"
(I worry about that guy...)

Enable dirty buffer detection in mg.
Emulate the emacs behavior: after suspend/resume, buffer switch,
or at save time, warn (prompt) the user if the file has been modified
on disk in the interim.
This has already saved my butt numerous times.
ok phessler

typos; ok jmc@
sys/dev/pci/pciide.c from naddy@

Fix a bug where inserting a file resulted in an incorrect
line-number count for a buffer (M-X insert-file, M-> to reproduce).
While here, fix a number of bugs with incorrect line numbers
after swap point-and-mark

Originally reported via debian's bug tracking system. Fix tested by
Han Boetes and Deanna Phillips.

spacing

Use S_IS* macros insted of masking with S_IF* flags. The latter may
have multiple bits set, which lead to surprising results. Spotted by
Paul Stoeber, more to come. ok millert@ pedro@ jaredy@ djm@

Rename the header line of a buffer to b_headp, from the remarkably
unintuitive b_linep. No binary change.

- use <dirent.h> not <sys/dir.h> (this appeases FreeBSD's annoying #warning)
- grab <sys/time.h> before <sys/resource.h> like getrusage(2) says
ok kjell

fix style

make // /~ path rewriting optional in adjustname() and use it everywhere
except for the command line specified files. ok kjell,cloder

correct one more comment

Do some KNF, clean up some unused junk that has lying around for 40-odd
cvs revisions, and fix a few comments to match reality

len = strlen(foo); if foo[len - 1]... will do bad things if len == 0.
Avoid, and rearrange a test so it can't (hypothetically) overflow.

lint love; ok kjell

test char against '\0', not 0

Clean up the ugly casted frees. In one case, this meant eliminating a nasty
struct/union/casting nightmare when building the list of names for filename
completion. In particular, be consistent about strduping and freeing
the list data.

Do some delinting of strl-type functions. Also, remove a superfluous
word in the undo-list.

do some silly de-linting

#ifdef NO_DIR can go too. From Han Boetes.

Kill the NO_BACKUP #ifdef

toast NO_DIRED #ifdef; ok kjell

greedy use of typedef struct was making code harder to read; ok kjell cloder

Better error checking of snprintfs. From Han Boetes.

Now that we have an editable minibuffer, revert parsing of
"//" and "/~" as '/' and '~' respectively. I know emacs does it,
but it is weird, and breaks things that foolishly use filenames
like "/tmp//crontab.xxxx".
Proddings from matthieu and deraadt. Error report from Bernd Ahlers.

spacing

When attempting to open a filename, walk backwards through minibuffer:
a> if you run into the beginning of the string, use the whole thing
b> if you run into a // combo, use everything starting from the second /
c> if you run into a /~ combo, use everything starting from the ~
i.e. do like emacs.
From (and for) Jason Wright

Fix string handling. LOGIN_NAME_MAX includes the NUL, so don't reserve extra
space. Fix usage of strlcpy. Correct test for terminating slash.

"looks good" deraadt@

more headers to please lint

move the dired routines into dired.c where they belong.
ok cloder@

Make dired buffer read-only by default.
Noticed by Han Boetes

Use dired mode automatically if file specified for loading is
a directory. Modified version of patch from Han Boetes.
ok cloder@

check if current line is null. fixes a core in dired.

Fix dired mode. Make deletions work, and keystrokes match emacs.
* dired-other-window should default to current buffer's path.
* Remove redundant code (from Han Boetes)
* Move initialization (and since we're here, fix mail-mode initialization
too)
* Remove redundant keymap in dired.c, and replace it with
the better one that was being ignored in keymap.c.

Add explicit public domain notices to all public domain files.
ok millert@, deraadt@

This is a no binary change which does:

- spelling, punctuation fixes
- variable declaration lineup
- use parentheses for return and sizeof
- K&R function declarations -> ANSI
- other minor code beautification

ok henning@

spacing

the realpath() was needed to do do relative to absolute path
conversion. revert my diff that takes it out, and instead, on
realpath() failure (like when you have no perms on pwd) just return the
unexpanded path. everything still fine with that, just tab completeion
and the like does not work (obviously).
this was mainly for "sudo mg /etc/something" from ~ where ~ is
nfs-mounted with root mapped to -2
and now, that case works as well as tab completion on insert-file etc.

fix mg's behaviour with regards to files on which we do not have
write access. diff originally from vincent@ and forgotten for
a while.

ok rohee@, "toss it in, i'll bitch if it doesn't work" henning@

make fbackupfile() return FALSE when the rename(2) from the
mkstemp()-derived name to the filename~ one fails
From: Han Boetes <han@mijncomputer.nl>

Safer backup file generation using mkstemp(); henning@ OK

no need to call realpath() in adjustname()
realpath returns an error if cwd is not writeable, and that is quite common
with NFS-mounted ~ and sudo mg /etc/something - annoying as hell. realpath
is not needed at all here, so zap it. vincent ok

when saving a file, check whether it is newline terminated and if it is not,
prompt the user and add the newline if he agrees.

tested by a lot of helpful tech@ guys, approved by henning

simplify the file writing loop

looked at by ian, tested on i386, macppc, sparc64

read(2) returns ssize_t, not size_t. spotted by Andrey Matveev.

typos from Jonathon Gray;

make dired work more like emacs. (well, make it _work_, and then add a
bunch of missing commands)

actually extract the filename off the ls line. makes dired *slightly*
more useful.

use NULL to indicate failure in functions that return pointers, not FALSE.
typo in comment

from Patrick Latifi (patrick.l@hermes.usherb.ca)

fix a small memleak on errors. noticed by pat latifi.

ok millert, fgsch

knf

replace the ugly and buggy adjustname function by a simple one using
simple APIs. makes mg not crash with 65k filenames...

ok art@

KNF

fix bad usage of strlcpy()'s return value. (made pointers point
beyond the boundaries of buffers)

ok deraadt@

don't use /bin/cp to copy files.

ok deraadt@

fix mem leak

ok deraadt@

use fchmod and fchown instead; vincent

Fixes the the following problems (from zyrnix)
1) Mg crashes with column width of 1, rows > 2
2) Mg panics on resizing due to incomplete write
3) Mg doesn't compile with STARTUPFILE defined
deraadt@ OK

Make these special buffers readonly by default. ok art@

* Move to ANSI function definitions.
* Add a whole lot of consts where I thought it made sense

no ok, but no objections either...

if file mode known, try to avoid a race.. i think. millert spotted a bug in my first draft

d_makename now takes a length parameter so we can remove the
strcpy call.

some KNF

* Replace unsafe strcpy and strcat calls to safe strlcpy and strlcat.
* Be a little bit more verbose about some errors
* Fix some memory leaks in fileio.c

ok deraadt@, art@

correct fork logic and rewrite for clarity, ok todd@

correct type on last arg to execl(); nordin@cse.ogi.edu

Change the filename completion code to:
- not use an unprotected strcat.
- Append a '/' when a file is a symlink to a directory.
(tab completion on '/sys' is now less annoying.)
- actually make some more sense.

Don't assume that map_table[0] is the fundamental map.

spaces

use strlcpy vs cpy+len, some other fixes

Get rid of unnecessary casts of NULL.

remove VOID. We're way beyond the point where this could build with an
ancient compiler.

* Let make_file_list always append a '/' at the end of directories.
This makes filename TAB-completion more useful, now you don't need to
write that '/' manually.
* random cleanups including using snprintf and memcmp where open-coded
equivalents were used, waitpid instead of 'while (wait() != pid);', etc.

Use opendir/readdir instead of open/read when making a file list.
This makes filename tab-completion work and takes dired a step further.

$OpenBSD$

Don't use rename to backup the file to a ~ file. We need to make a
copy instead so the original is edited in place for things like
cron and vipw.

$HOME paranoia: never use getenv("HOME") w/o checking for NULL and non-zero

The start of KNF + -Wall. The code has been run through indent but
needs hand fixup. I stopped at keymap.c...

Remove a now-bogus comment
Use pid_t
Use vfork(), not fork() when sensible

We don't have silly SYSV 14-character file name limits. Taken from
the BSD fileio.c. Note that this module is rife with PATH_MAX overflow
possibilities.

Use uid_t, gid_t and mode_t where applicable.
Pull in changes from the BSD-specific sysdef.h and fileio.c:
kill bogus unlinkdir() and rename() kludges since we have rmdir(2) and rename(2)
define SYMBLINK since we have symlinks
TODO: merge BSD spawn.c and ttyio.c

further pruning

make it minimally compile; millert

initial import of mg2a

Delete obsolete /* ARGSUSED */ lint comments.

ok miod@ millert@

Put the hardcoded '\n' character which is found throughout mg into a
buffer specific variable. The diff should not produce any behavourial
changes in mg.

Allow the user to specify a path to the mg startup file on the command line.

This diff moves the check for a newline at the end of a buffer (when
saving it) earlier. This means data doesn't 'disappear' from the file
on disk while the user answers 'yes' or 'no'.

mg tries to write backups to the wrong directory when run under a
different effective user, i.e. when invoced via su and
backup-to-home-directory is enabled.

Problem pointed out and diff provied by Lucas Gabriel Vuotto
<lvuotto92 () gmail ! com>, thanks!

Subsequently slacked on for nearly a year by yours truly.

Then remembered when Han Boetes <hboetes () gmail ! com> came up with
a similar diff because of a problem report by Mark Willson where it
turned out that getlogin(2) is not very portable.

OK tb

fix -Wshadow warnings.
Found by bcallah, thanks!

strncmp is a more reasonable way to check the beginning of a string.

these programs probably do not need to use TMPDIR. ok florian

Use fstatat() instead of crafting a filename to use with stat()

ok millert@

ensure the backup file has the same mtime as the original file, this is in line
with emacs' behaviour.

from Harald Dunkel via Han Boetes
ok lum@

Clean up the includes in mg.
This does the following:
Moves all POSIX headers from sysdef.h into the individual .c files so that
each file now only includes what it needs. All headers are properly sorted.
Moves the remainder of sysdef.h to other files (mostly def.h) and deletes
sysdef.h now that it's no longer contains anything.
Tweak a comment that references sysdef.h so that it no longer does that.
ok florian@

Stop using <sys/param.h>; replace MAXPATHLEN with PATH_MAX, stop using MAX(),
and pull in <limits.h> for *_MAX constants.

inspired on a diff from Kamil Rytarowski (n54 (at) gmx.com)
ok bcallah@

Add some missing dobeeps.
ok florian@

and and -> and

- move prototype of expandtilder() to def.h; needed by an upcoming diff.

Fix an outstanding debian bug: #658539

"looks good" kjell

- add a pathnames.h
- some whitespace cleanup

ok lum@

Fixes a problem where if you try to open a file that doesn't exist and has
a name longer than LOGIN_NAME_MAX and also has a tilde at the front
e.g:

$ mg ~01234567890123456789012345678901

mg will give a "Login name too long" instead of opening a new buffer
named ~01234567890123456789012345678901

Allow mg to save backup files to a users home directory.
Suggestions from eric@ and Sunil Nimmagadda.
Remarks deraadt@.

Change error message to reflect the correct filename: nname is 'New
name' as opposed to tname 'Temp name'

Remove static FILE pointer used for handling files in fileio.c. Pass
by reference instead. This allows the mg startup file to open other
files without unexpected things happening.

Discussed with Sunil Nimmagadda.

Change how mg handles files beginning with a '~'.
Bring behaviour more into line with emacs.

Remove the conditional directives NO_MACRO and NO_STARTUP.
They have not compiled for numerous years.
ok kjell@ millert@

Close an opendir(). From Igor Zinovik.

On a file write fail:

1. return an error value
2. show an error message

From Loganaden Velvindron with suggestion from millert@

It volates style(9), but in mg, #include"def.h" goes first. 'twas the way it was built. no binary change here. confirmed by lum@, tested by Henri Kemppainen

Change len+memcmp check to a strncmp for prefix testing on
filename completion. Avoids a recurring portability headache.
ok lum@
"concerns assuaged to the soothing sounds of Martin Denny blambert@"
(I worry about that guy...)

Enable dirty buffer detection in mg.
Emulate the emacs behavior: after suspend/resume, buffer switch,
or at save time, warn (prompt) the user if the file has been modified
on disk in the interim.
This has already saved my butt numerous times.
ok phessler

typos; ok jmc@
sys/dev/pci/pciide.c from naddy@

Fix a bug where inserting a file resulted in an incorrect
line-number count for a buffer (M-X insert-file, M-> to reproduce).
While here, fix a number of bugs with incorrect line numbers
after swap point-and-mark

Originally reported via debian's bug tracking system. Fix tested by
Han Boetes and Deanna Phillips.

spacing

Use S_IS* macros insted of masking with S_IF* flags. The latter may
have multiple bits set, which lead to surprising results. Spotted by
Paul Stoeber, more to come. ok millert@ pedro@ jaredy@ djm@

Rename the header line of a buffer to b_headp, from the remarkably
unintuitive b_linep. No binary change.

- use <dirent.h> not <sys/dir.h> (this appeases FreeBSD's annoying #warning)
- grab <sys/time.h> before <sys/resource.h> like getrusage(2) says
ok kjell

fix style

make // /~ path rewriting optional in adjustname() and use it everywhere
except for the command line specified files. ok kjell,cloder

correct one more comment

Do some KNF, clean up some unused junk that has lying around for 40-odd
cvs revisions, and fix a few comments to match reality

len = strlen(foo); if foo[len - 1]... will do bad things if len == 0.
Avoid, and rearrange a test so it can't (hypothetically) overflow.

lint love; ok kjell

test char against '\0', not 0

Clean up the ugly casted frees. In one case, this meant eliminating a nasty
struct/union/casting nightmare when building the list of names for filename
completion. In particular, be consistent about strduping and freeing
the list data.

Do some delinting of strl-type functions. Also, remove a superfluous
word in the undo-list.

do some silly de-linting

#ifdef NO_DIR can go too. From Han Boetes.

Kill the NO_BACKUP #ifdef

toast NO_DIRED #ifdef; ok kjell

greedy use of typedef struct was making code harder to read; ok kjell cloder

Better error checking of snprintfs. From Han Boetes.

Now that we have an editable minibuffer, revert parsing of
"//" and "/~" as '/' and '~' respectively. I know emacs does it,
but it is weird, and breaks things that foolishly use filenames
like "/tmp//crontab.xxxx".
Proddings from matthieu and deraadt. Error report from Bernd Ahlers.

spacing

When attempting to open a filename, walk backwards through minibuffer:
a> if you run into the beginning of the string, use the whole thing
b> if you run into a // combo, use everything starting from the second /
c> if you run into a /~ combo, use everything starting from the ~
i.e. do like emacs.
From (and for) Jason Wright

Fix string handling. LOGIN_NAME_MAX includes the NUL, so don't reserve extra
space. Fix usage of strlcpy. Correct test for terminating slash.

"looks good" deraadt@

more headers to please lint

move the dired routines into dired.c where they belong.
ok cloder@

Make dired buffer read-only by default.
Noticed by Han Boetes

Use dired mode automatically if file specified for loading is
a directory. Modified version of patch from Han Boetes.
ok cloder@

check if current line is null. fixes a core in dired.

Fix dired mode. Make deletions work, and keystrokes match emacs.
* dired-other-window should default to current buffer's path.
* Remove redundant code (from Han Boetes)
* Move initialization (and since we're here, fix mail-mode initialization
too)
* Remove redundant keymap in dired.c, and replace it with
the better one that was being ignored in keymap.c.

Add explicit public domain notices to all public domain files.
ok millert@, deraadt@

This is a no binary change which does:

- spelling, punctuation fixes
- variable declaration lineup
- use parentheses for return and sizeof
- K&R function declarations -> ANSI
- other minor code beautification

ok henning@

spacing

the realpath() was needed to do do relative to absolute path
conversion. revert my diff that takes it out, and instead, on
realpath() failure (like when you have no perms on pwd) just return the
unexpanded path. everything still fine with that, just tab completeion
and the like does not work (obviously).
this was mainly for "sudo mg /etc/something" from ~ where ~ is
nfs-mounted with root mapped to -2
and now, that case works as well as tab completion on insert-file etc.

fix mg's behaviour with regards to files on which we do not have
write access. diff originally from vincent@ and forgotten for
a while.

ok rohee@, "toss it in, i'll bitch if it doesn't work" henning@

make fbackupfile() return FALSE when the rename(2) from the
mkstemp()-derived name to the filename~ one fails
From: Han Boetes <han@mijncomputer.nl>

Safer backup file generation using mkstemp(); henning@ OK

no need to call realpath() in adjustname()
realpath returns an error if cwd is not writeable, and that is quite common
with NFS-mounted ~ and sudo mg /etc/something - annoying as hell. realpath
is not needed at all here, so zap it. vincent ok

when saving a file, check whether it is newline terminated and if it is not,
prompt the user and add the newline if he agrees.

tested by a lot of helpful tech@ guys, approved by henning

simplify the file writing loop

looked at by ian, tested on i386, macppc, sparc64

read(2) returns ssize_t, not size_t. spotted by Andrey Matveev.

typos from Jonathon Gray;

make dired work more like emacs. (well, make it _work_, and then add a
bunch of missing commands)

actually extract the filename off the ls line. makes dired *slightly*
more useful.

use NULL to indicate failure in functions that return pointers, not FALSE.
typo in comment

from Patrick Latifi (patrick.l@hermes.usherb.ca)

fix a small memleak on errors. noticed by pat latifi.

ok millert, fgsch

knf

replace the ugly and buggy adjustname function by a simple one using
simple APIs. makes mg not crash with 65k filenames...

ok art@

KNF

fix bad usage of strlcpy()'s return value. (made pointers point
beyond the boundaries of buffers)

ok deraadt@

don't use /bin/cp to copy files.

ok deraadt@

fix mem leak

ok deraadt@

use fchmod and fchown instead; vincent

Fixes the the following problems (from zyrnix)
1) Mg crashes with column width of 1, rows > 2
2) Mg panics on resizing due to incomplete write
3) Mg doesn't compile with STARTUPFILE defined
deraadt@ OK

Make these special buffers readonly by default. ok art@

* Move to ANSI function definitions.
* Add a whole lot of consts where I thought it made sense

no ok, but no objections either...

if file mode known, try to avoid a race.. i think. millert spotted a bug in my first draft

d_makename now takes a length parameter so we can remove the
strcpy call.

some KNF

* Replace unsafe strcpy and strcat calls to safe strlcpy and strlcat.
* Be a little bit more verbose about some errors
* Fix some memory leaks in fileio.c

ok deraadt@, art@

correct fork logic and rewrite for clarity, ok todd@

correct type on last arg to execl(); nordin@cse.ogi.edu

Change the filename completion code to:
- not use an unprotected strcat.
- Append a '/' when a file is a symlink to a directory.
(tab completion on '/sys' is now less annoying.)
- actually make some more sense.

Don't assume that map_table[0] is the fundamental map.

spaces

use strlcpy vs cpy+len, some other fixes

Get rid of unnecessary casts of NULL.

remove VOID. We're way beyond the point where this could build with an
ancient compiler.

* Let make_file_list always append a '/' at the end of directories.
This makes filename TAB-completion more useful, now you don't need to
write that '/' manually.
* random cleanups including using snprintf and memcmp where open-coded
equivalents were used, waitpid instead of 'while (wait() != pid);', etc.

Use opendir/readdir instead of open/read when making a file list.
This makes filename tab-completion work and takes dired a step further.

$OpenBSD$

Don't use rename to backup the file to a ~ file. We need to make a
copy instead so the original is edited in place for things like
cron and vipw.

$HOME paranoia: never use getenv("HOME") w/o checking for NULL and non-zero

The start of KNF + -Wall. The code has been run through indent but
needs hand fixup. I stopped at keymap.c...

Remove a now-bogus comment
Use pid_t
Use vfork(), not fork() when sensible

We don't have silly SYSV 14-character file name limits. Taken from
the BSD fileio.c. Note that this module is rife with PATH_MAX overflow
possibilities.

Use uid_t, gid_t and mode_t where applicable.
Pull in changes from the BSD-specific sysdef.h and fileio.c:
kill bogus unlinkdir() and rename() kludges since we have rmdir(2) and rename(2)
define SYMBLINK since we have symlinks
TODO: merge BSD spawn.c and ttyio.c

further pruning

make it minimally compile; millert

initial import of mg2a

Put the hardcoded '\n' character which is found throughout mg into a
buffer specific variable. The diff should not produce any behavourial
changes in mg.

Allow the user to specify a path to the mg startup file on the command line.

This diff moves the check for a newline at the end of a buffer (when
saving it) earlier. This means data doesn't 'disappear' from the file
on disk while the user answers 'yes' or 'no'.

mg tries to write backups to the wrong directory when run under a
different effective user, i.e. when invoced via su and
backup-to-home-directory is enabled.

Problem pointed out and diff provied by Lucas Gabriel Vuotto
<lvuotto92 () gmail ! com>, thanks!

Subsequently slacked on for nearly a year by yours truly.

Then remembered when Han Boetes <hboetes () gmail ! com> came up with
a similar diff because of a problem report by Mark Willson where it
turned out that getlogin(2) is not very portable.

OK tb

fix -Wshadow warnings.
Found by bcallah, thanks!

strncmp is a more reasonable way to check the beginning of a string.

these programs probably do not need to use TMPDIR. ok florian

Use fstatat() instead of crafting a filename to use with stat()

ok millert@

ensure the backup file has the same mtime as the original file, this is in line
with emacs' behaviour.

from Harald Dunkel via Han Boetes
ok lum@

Clean up the includes in mg.
This does the following:
Moves all POSIX headers from sysdef.h into the individual .c files so that
each file now only includes what it needs. All headers are properly sorted.
Moves the remainder of sysdef.h to other files (mostly def.h) and deletes
sysdef.h now that it's no longer contains anything.
Tweak a comment that references sysdef.h so that it no longer does that.
ok florian@

Stop using <sys/param.h>; replace MAXPATHLEN with PATH_MAX, stop using MAX(),
and pull in <limits.h> for *_MAX constants.

inspired on a diff from Kamil Rytarowski (n54 (at) gmx.com)
ok bcallah@

Add some missing dobeeps.
ok florian@

and and -> and

- move prototype of expandtilder() to def.h; needed by an upcoming diff.

Fix an outstanding debian bug: #658539

"looks good" kjell

- add a pathnames.h
- some whitespace cleanup

ok lum@

Fixes a problem where if you try to open a file that doesn't exist and has
a name longer than LOGIN_NAME_MAX and also has a tilde at the front
e.g:

$ mg ~01234567890123456789012345678901

mg will give a "Login name too long" instead of opening a new buffer
named ~01234567890123456789012345678901

Allow mg to save backup files to a users home directory.
Suggestions from eric@ and Sunil Nimmagadda.
Remarks deraadt@.

Change error message to reflect the correct filename: nname is 'New
name' as opposed to tname 'Temp name'

Remove static FILE pointer used for handling files in fileio.c. Pass
by reference instead. This allows the mg startup file to open other
files without unexpected things happening.

Discussed with Sunil Nimmagadda.

Change how mg handles files beginning with a '~'.
Bring behaviour more into line with emacs.

Remove the conditional directives NO_MACRO and NO_STARTUP.
They have not compiled for numerous years.
ok kjell@ millert@

Close an opendir(). From Igor Zinovik.

On a file write fail:

1. return an error value
2. show an error message

From Loganaden Velvindron with suggestion from millert@

It volates style(9), but in mg, #include"def.h" goes first. 'twas the way it was built. no binary change here. confirmed by lum@, tested by Henri Kemppainen

Change len+memcmp check to a strncmp for prefix testing on
filename completion. Avoids a recurring portability headache.
ok lum@
"concerns assuaged to the soothing sounds of Martin Denny blambert@"
(I worry about that guy...)

Enable dirty buffer detection in mg.
Emulate the emacs behavior: after suspend/resume, buffer switch,
or at save time, warn (prompt) the user if the file has been modified
on disk in the interim.
This has already saved my butt numerous times.
ok phessler

typos; ok jmc@
sys/dev/pci/pciide.c from naddy@

Fix a bug where inserting a file resulted in an incorrect
line-number count for a buffer (M-X insert-file, M-> to reproduce).
While here, fix a number of bugs with incorrect line numbers
after swap point-and-mark

Originally reported via debian's bug tracking system. Fix tested by
Han Boetes and Deanna Phillips.

spacing

Use S_IS* macros insted of masking with S_IF* flags. The latter may
have multiple bits set, which lead to surprising results. Spotted by
Paul Stoeber, more to come. ok millert@ pedro@ jaredy@ djm@

Rename the header line of a buffer to b_headp, from the remarkably
unintuitive b_linep. No binary change.

- use <dirent.h> not <sys/dir.h> (this appeases FreeBSD's annoying #warning)
- grab <sys/time.h> before <sys/resource.h> like getrusage(2) says
ok kjell

fix style

make // /~ path rewriting optional in adjustname() and use it everywhere
except for the command line specified files. ok kjell,cloder

correct one more comment

Do some KNF, clean up some unused junk that has lying around for 40-odd
cvs revisions, and fix a few comments to match reality

len = strlen(foo); if foo[len - 1]... will do bad things if len == 0.
Avoid, and rearrange a test so it can't (hypothetically) overflow.

lint love; ok kjell

test char against '\0', not 0

Clean up the ugly casted frees. In one case, this meant eliminating a nasty
struct/union/casting nightmare when building the list of names for filename
completion. In particular, be consistent about strduping and freeing
the list data.

Do some delinting of strl-type functions. Also, remove a superfluous
word in the undo-list.

do some silly de-linting

#ifdef NO_DIR can go too. From Han Boetes.

Kill the NO_BACKUP #ifdef

toast NO_DIRED #ifdef; ok kjell

greedy use of typedef struct was making code harder to read; ok kjell cloder

Better error checking of snprintfs. From Han Boetes.

Now that we have an editable minibuffer, revert parsing of
"//" and "/~" as '/' and '~' respectively. I know emacs does it,
but it is weird, and breaks things that foolishly use filenames
like "/tmp//crontab.xxxx".
Proddings from matthieu and deraadt. Error report from Bernd Ahlers.

spacing

When attempting to open a filename, walk backwards through minibuffer:
a> if you run into the beginning of the string, use the whole thing
b> if you run into a // combo, use everything starting from the second /
c> if you run into a /~ combo, use everything starting from the ~
i.e. do like emacs.
From (and for) Jason Wright

Fix string handling. LOGIN_NAME_MAX includes the NUL, so don't reserve extra
space. Fix usage of strlcpy. Correct test for terminating slash.

"looks good" deraadt@

more headers to please lint

move the dired routines into dired.c where they belong.
ok cloder@

Make dired buffer read-only by default.
Noticed by Han Boetes

Use dired mode automatically if file specified for loading is
a directory. Modified version of patch from Han Boetes.
ok cloder@

check if current line is null. fixes a core in dired.

Fix dired mode. Make deletions work, and keystrokes match emacs.
* dired-other-window should default to current buffer's path.
* Remove redundant code (from Han Boetes)
* Move initialization (and since we're here, fix mail-mode initialization
too)
* Remove redundant keymap in dired.c, and replace it with
the better one that was being ignored in keymap.c.

Add explicit public domain notices to all public domain files.
ok millert@, deraadt@

This is a no binary change which does:

- spelling, punctuation fixes
- variable declaration lineup
- use parentheses for return and sizeof
- K&R function declarations -> ANSI
- other minor code beautification

ok henning@

spacing

the realpath() was needed to do do relative to absolute path
conversion. revert my diff that takes it out, and instead, on
realpath() failure (like when you have no perms on pwd) just return the
unexpanded path. everything still fine with that, just tab completeion
and the like does not work (obviously).
this was mainly for "sudo mg /etc/something" from ~ where ~ is
nfs-mounted with root mapped to -2
and now, that case works as well as tab completion on insert-file etc.

fix mg's behaviour with regards to files on which we do not have
write access. diff originally from vincent@ and forgotten for
a while.

ok rohee@, "toss it in, i'll bitch if it doesn't work" henning@

make fbackupfile() return FALSE when the rename(2) from the
mkstemp()-derived name to the filename~ one fails
From: Han Boetes <han@mijncomputer.nl>

Safer backup file generation using mkstemp(); henning@ OK

no need to call realpath() in adjustname()
realpath returns an error if cwd is not writeable, and that is quite common
with NFS-mounted ~ and sudo mg /etc/something - annoying as hell. realpath
is not needed at all here, so zap it. vincent ok

when saving a file, check whether it is newline terminated and if it is not,
prompt the user and add the newline if he agrees.

tested by a lot of helpful tech@ guys, approved by henning

simplify the file writing loop

looked at by ian, tested on i386, macppc, sparc64

read(2) returns ssize_t, not size_t. spotted by Andrey Matveev.

typos from Jonathon Gray;

make dired work more like emacs. (well, make it _work_, and then add a
bunch of missing commands)

actually extract the filename off the ls line. makes dired *slightly*
more useful.

use NULL to indicate failure in functions that return pointers, not FALSE.
typo in comment

from Patrick Latifi (patrick.l@hermes.usherb.ca)

fix a small memleak on errors. noticed by pat latifi.

ok millert, fgsch

knf

replace the ugly and buggy adjustname function by a simple one using
simple APIs. makes mg not crash with 65k filenames...

ok art@

KNF

fix bad usage of strlcpy()'s return value. (made pointers point
beyond the boundaries of buffers)

ok deraadt@

don't use /bin/cp to copy files.

ok deraadt@

fix mem leak

ok deraadt@

use fchmod and fchown instead; vincent

Fixes the the following problems (from zyrnix)
1) Mg crashes with column width of 1, rows > 2
2) Mg panics on resizing due to incomplete write
3) Mg doesn't compile with STARTUPFILE defined
deraadt@ OK

Make these special buffers readonly by default. ok art@

* Move to ANSI function definitions.
* Add a whole lot of consts where I thought it made sense

no ok, but no objections either...

if file mode known, try to avoid a race.. i think. millert spotted a bug in my first draft

d_makename now takes a length parameter so we can remove the
strcpy call.

some KNF

* Replace unsafe strcpy and strcat calls to safe strlcpy and strlcat.
* Be a little bit more verbose about some errors
* Fix some memory leaks in fileio.c

ok deraadt@, art@

correct fork logic and rewrite for clarity, ok todd@

correct type on last arg to execl(); nordin@cse.ogi.edu

Change the filename completion code to:
- not use an unprotected strcat.
- Append a '/' when a file is a symlink to a directory.
(tab completion on '/sys' is now less annoying.)
- actually make some more sense.

Don't assume that map_table[0] is the fundamental map.

spaces

use strlcpy vs cpy+len, some other fixes

Get rid of unnecessary casts of NULL.

remove VOID. We're way beyond the point where this could build with an
ancient compiler.

* Let make_file_list always append a '/' at the end of directories.
This makes filename TAB-completion more useful, now you don't need to
write that '/' manually.
* random cleanups including using snprintf and memcmp where open-coded
equivalents were used, waitpid instead of 'while (wait() != pid);', etc.

Use opendir/readdir instead of open/read when making a file list.
This makes filename tab-completion work and takes dired a step further.

$OpenBSD$

Don't use rename to backup the file to a ~ file. We need to make a
copy instead so the original is edited in place for things like
cron and vipw.

$HOME paranoia: never use getenv("HOME") w/o checking for NULL and non-zero

The start of KNF + -Wall. The code has been run through indent but
needs hand fixup. I stopped at keymap.c...

Remove a now-bogus comment
Use pid_t
Use vfork(), not fork() when sensible

We don't have silly SYSV 14-character file name limits. Taken from
the BSD fileio.c. Note that this module is rife with PATH_MAX overflow
possibilities.

Use uid_t, gid_t and mode_t where applicable.
Pull in changes from the BSD-specific sysdef.h and fileio.c:
kill bogus unlinkdir() and rename() kludges since we have rmdir(2) and rename(2)
define SYMBLINK since we have symlinks
TODO: merge BSD spawn.c and ttyio.c

further pruning

make it minimally compile; millert

initial import of mg2a

Allow the user to specify a path to the mg startup file on the command line.

This diff moves the check for a newline at the end of a buffer (when
saving it) earlier. This means data doesn't 'disappear' from the file
on disk while the user answers 'yes' or 'no'.