| Age | Commit message (Collapse) | Author |
|---|
|
|
|
FD_CLOEXEC is much more appropriate here.
|
|
|
|
|
|
|
|
* signatures no longer deal with zcat. Instead, we sign the gzip file
itself (stripped of the signature part of the header, of course). Thanks
Angelos. Niels seems to think passing the header itself to sign is not
a problem, even though no-one cares about checking it ?
* gzip header handling revamped: can write to memory. Will eliminate some
pipes later on. Can stack signatures.
* taken out specific signature schemes (e.g., pgp and sha1). Code is now
signature scheme independent, mostly, and writes with client data from
memory, e.g., check.c can invoke several checks in parallel without needing
to fork.
* need the full set of popen-like functionalities (keep track of opened
file descriptors to avoid passing them down to children)
* remove simple_check.c, functionality absorbed elsewhere.
To do:
* re-check message output and what to do with unsigned/unchecked/verified
packages,
* check pkg_add implementation and remove extra-pipe in asynchronous
checking,
* control over what to do when several signatures are stacked... Simple
way would be to disable that for now (possibility for release)
* get the code through a linter again.
|
|
file.
|
|
|
|
Requires pgp for now, though the general scheme is generic.
|
|
- disable FreeBSD-specific packages organization per category
|
|
Removes an annoying limit on PLIST lengths.
|
|
|
|
* except when it is escaped with a `\' at the end of the line
- fix remaining .Nm usage as well
- this is from a patch I received from kwesterback@home.com, who has been
working on some scripts for fixing formatting errors in mdoc'd man pages
Ok, so there could be a cost/benefit debate with this commit, but since I have
the patch we might as well commit it...
|
|
|
|
|
|
(This could have been done by using pax -n, but since pkg_* is so badly
written, it's harder to hunt down all the code using tar and fix it to
use the right options.
If this code were written properly, there would be one clean entry point
to processing archives, and it would be trivial to switch archivers...)
|
|
|
|
- use the new form of chown, with user:group instead of user.group.
|
|
|
|
|
|
|
|
|
|
|
|
Pending a check and a rewriting, as the documentation is misleading.
It seems to imply it looks thru installed packages first, whereas
code tells pkg_info looks thru package files first.
|
|
now tries to complete the package name with a version number while
scanning the installed list of packages.
Not foolproof. The pkg_* tools are atrocious anyway.
|
|
|
|
|
|
|
|
- warn when COMMENT > 60 characters.
*** In one month we turn this into an error -> fix packages files ***
|
|
- use snprintf in place of sprintf
- code cleanup
- Package -> package_t, PackingList -> plist_t
Also: remove files that haven't been linked in a while
Pass -q to mtree so it is quiet in the presence of symlinks
|
|
New features include md5 hash so pkg_delete won't remove files that have
changed and the ability to define conflicting packages, e.g. you can't
install both mh and nmh. The ports tree will have to be updated to take
advantage of this.
Let me know of any problems, real or imagined :-)
|
|
occurrences squashed as well.
|
|
|
|
|
|
|
|
|
|
stdin twice. Note: it assumes stdin is a compressed tar file.
|
|
|
|
|
|
|
|
|
|
Quote filenames passed to different extract commands preventing
certain security risks.
|
|
to the fact that GNU tar retatins the bits when run as root, but paxtar does
not, and we use the latter.
|
|
done.
|
|
These should be checked in all but one case anyway, but you never know.
|
|
pkg_info by itself acts as if -a -I were given, and
when no flags are passed with package names, -I is implied.
|
|
Also, minor KNF whlle here.
|
|
|
|
committing :-) ... i hope i got it right now
|
|
|
