Age | Commit message (Collapse) | Author |
|
Paths could contain characters like # (fragment), '?', control-characters, etc.
|
|
Like ", which would unquote the attribute value. Crazy but true.
|
|
Reference:
https://libgit2.org/libgit2/#HEAD/group/revwalk/git_revwalk_simplify_first_parent
Noticed on merge commits on:
https://git.simple-cc.org/scc/
Reported by quinq, thanks!
|
|
|
|
This can happen when there is no newline at end of file in the diff which is
served by libgit2 as:
"\n\ No newline at end of file\n".
|
|
|
|
With feedback from adc, thanks!
|
|
|
|
git_blob_rawsize now returns with git_object_size_t (unsigned). This was
git_off_t (signed).
In my current version 1.1.0:
types.h:typedef uint64_t git_object_size_t;
v0.28.5:
https://libgit2.org/libgit2/#v0.28.5/group/blob/git_blob_rawsize
changed from v0.99 onwards:
https://libgit2.org/libgit2/#v0.99.0/group/blob/git_blob_rawsize
Fix: use size_t to remain compatible (with a possible warning in older
versions), since git_object_size_t is a new defined type.
This assumes size_t is atleast uint32_t / uint64_t size.
Adapted from a patch by Augustin Fabre <augustin@augfab.fr>, thanks!
|
|
|
|
Patch by Augustin Fabre <augustin@augfab.fr>
|
|
|
|
by Augustin Fabre <augustin@augfab.fr>
|
|
|
|
from Augustin Fabre <augustin@augfab.fr>
|
|
non-tag references were not freed.
|
|
git_submodule_lookup does not work without a working tree [1], so the
current approach fails to recognize any submodules in bare repos.
Instead, notice that
$ git ls-tree HEAD
lists any submodules as commit objects regardless of a working tree.
This is the only instance commit object is used in a tree, so we will
use this to check for submodules.
[1]: https://github.com/libgit2/libgit2/pull/4305/files
|
|
|
|
A separate Atom feed is helpful to ports maintainers to monitor new
tags/releases.
|
|
In general version tags are done in chronological order, so this will have a
better sorting for tagged (versioned) releases.
Request from Caltlgin Stsodaat and others, thanks!
|
|
|
|
|
|
The unveil() system call first appeared in OpenBSD 6.4.
For stagit it has the following properties now:
- stagit-index: only read-access to the file-system for the specified
directories/repositories.
- stagit: read-access to the specified directory/repository.
read-write and creation access to the current directory for the output files.
read-write and creation access to the specified cache file.
|
|
- escape paths in diff line.
- escape path in anchor href attribute in files.html
|
|
POSIX says:
"If an output error was encountered, these functions shall return a negative
value and set errno to indicate the error."
|
|
|
|
libgit2 commit:
https://github.com/libgit2/libgit2/commit/a27a4de6f8003961d38958893c6c637395c7cc04
|
|
|
|
this prevents showing files as renames when most (but not all) of the file was
changed.
|
|
|
|
- for license: LICENSE, LICENSE.md, COPYING.
- for readme: README, README.md.
|
|
|
|
|
|
|
|
|
|
|
|
this is because libgit uses curl which uses some initialization
checks, namely an IPv6 check which creates a socket.
|
|
... also clear all fields in the structure on failure.
This is not as big an optimization as stagit-gopher, because the diffstat is
displayed in the log, but the difference is still measurable.
|
|
|
|
improve checks for when head is NULL (empty repo)
|
|
|
|
|
|
|
|
|
|
|
|
make sure not partial (invalid) utf-8 sequences are written. The old logic
truncated by bytelen was wrong. It is too complex to implement it in a
correct way with git: it is not assured a commit message is UTF-8 encoded.
remove showlinecount and summarylen variables.
|
|
- reorder: unnecessary allocation when git_patch_from_diff fails.
- no need to free patch object when git_patch_from_diff fails.
|
|
... now it is done.
|
|
using len instead of the loop 'i' is slightly more clear
|
|
else TABs wont render properly (interpreted as part of HTML whitespace)
at the first TAB of the line.
|