Skip to content

mingw: handle absolute paths in expand_user_path()

Johannes Schindelin via GitGitGadget git 2 versions – In review

Get series with b4
b4 am [email protected]

Versions newest first · linked by author and title

  1. v3 Latest5 patches 9 repliesIn review
  2. v22 patches 1 replySuperseded

Changes in v3 from the cover letter

* Adjusted to Junio's preference %(...) over <...>. Of course, the preferred preference has the disadvantage of actually being allowed in Win32 filenames, but then, the workaround is as easy as on non-Windows platforms.
* Since our convention of %(...) interpolation does not involve uppercase keywords, I now use a lowercase one.
* Since this keyword is interpolated to the compiled-in prefix if built without runtime prefix support, I dropped the runtime part of the keyword.
* Renamed the expand_user_path() to interpolate_path(), to remove the distraction as to the implementation detail which things get to be interpolated (because we extend it to interpolate more than just a home directory, which might well be unclear from the former name, anyway).
* Adjusted the code comment above the interpolate_path() to remove a stale part, clarify another part, and to extend it to talk about the prefix expansion, too.

Patches in v3 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/5tests: exercise the RUNTIME_PREFIX featureMakefile t/t0060-path-utils.sh+23 -00 replies
2/5expand_user_path(): remove stale part of the commentpath.c+2 -30 replies
3/5expand_user_path(): clarify the role of the `real_home` parameterpath.c+1 -10 replies
4/5Use a better name for the function interpolating pathsbuiltin/credential-cache.c builtin/credential-store.c builtin/gc.c +4 more+11 -118 replies
5/5interpolate_path(): allow specifying paths relative to the runtime prefixDocumentation/config.txt path.c t/t0060-path-utils.sh+25 -00 replies

Reviews trailers per patch

No Reviewed-by, Acked-by or Tested-by trailers for v3 yet.

Read the v3 thread

Cover letter [PATCH v3 0/5] mingw: handle absolute paths in expand_user_path()

In Git for Windows, we ran with a patch "in production" for quite a while where paths starting with a slash (i.e. looking like Unix paths, not like Windows paths) were interpreted as being relative to the runtime prefix, when expanded via expand_user_path().

This was sent to the Git mailing list as a discussion starter, and it was pointed out that this is neither portable nor unambiguous.

After the dust settled, I thought about the presented ideas for a while (quite a while...), ended up with a solution, then adapted it to Junio's
preference: any path starting with %(prefix)/ is expanded. This is ambiguous because it could be a valid path. But then, it is unlikely, and if someone really wants to specify such a path, it is easy to slap a ./ in front and they're done.

Changes since v2:

* Adjusted to Junio's preference %(...) over <...>. Of course, the preferred preference has the disadvantage of actually being allowed in Win32 filenames, but then, the workaround is as easy as on non-Windows platforms.
* Since our convention of %(...) interpolation does not involve uppercase keywords, I now use a lowercase one.
* Since this keyword is interpolated to the compiled-in prefix if built without runtime prefix support, I dropped the runtime part of the keyword.
* Renamed the expand_user_path() to interpolate_path(), to remove the distraction as to the implementation detail which things get to be interpolated (because we extend it to interpolate more than just a home directory, which might well be unclear from the former name, anyway).
* Adjusted the code comment above the interpolate_path() to remove a stale part, clarify another part, and to extend it to talk about the prefix expansion, too.

Changes since v1:

* Included a test for the RUNTIME_PREFIX that I had meant to send for ages already, and based on which...
* A test case was added to verify that this actually works as intended
* It is no longer Windows-specific
* I added some documentation

Johannes Schindelin (5):
tests: exercise the RUNTIME_PREFIX feature expand_user_path(): remove stale part of the comment expand_user_path(): clarify the role of the real_home parameter Use a better name for the function interpolating paths interpolate_path(): allow specifying paths relative to the runtime
prefix

 Documentation/config.txt   |  9 +++++++++
 Makefile                   |  5 +++++
 builtin/credential-cache.c |  2 +-
 builtin/credential-store.c |  2 +-
 builtin/gc.c               |  2 +-
 cache.h                    |  2 +-
 config.c                   |  8 ++++----
 path.c                     | 19 +++++++++++++------
 sequencer.c                |  2 +-
 t/t0060-path-utils.sh      | 26 ++++++++++++++++++++++++++
 10 files changed, 62 insertions(+), 15 deletions(-)

base-commit: ebf3c04b262aa27fbb97f8a0156c2347fecafafb
Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-66%2Fdscho%2Fmingw-expand-absolute-user-path-v3
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-66/dscho/mingw-expand-absolute-user-path-v3
Pull-Request: https://github.com/gitgitgadget/git/pull/66

Range-diff vs v2:

 1:  cc8f09baba9 = 1:  cc8f09baba9 tests: exercise the RUNTIME_PREFIX feature
 -:  ----------- > 2:  a4ff3a461bc expand_user_path(): remove stale part of the comment
 -:  ----------- > 3:  7b79ba66dd0 expand_user_path(): clarify the role of the `real_home` parameter
 -:  ----------- > 4:  19fd9c3c803 Use a better name for the function interpolating paths
 2:  66df56f5db0 ! 5:  d286583082e expand_user_path(): support specifying paths relative to the runtime prefix

@@ Metadata
Author: Johannes Schindelin <[email protected]>

## Commit message ##
- expand_user_path(): support specifying paths relative to the runtime prefix
+ interpolate_path(): allow specifying paths relative to the runtime prefix

Ever since Git learned to detect its install location at runtime, there was the slightly awkward problem that it was impossible to specify paths
@@ Commit message
they are interpreted as absolute paths in the same drive as the current directory.

- After a lengthy discussion, and a way lengthier time to mull over the
- problem and its best solution, we decided to introduce support for the
- magic sequence <RUNTIME-PREFIX>/. If a path starts with this, the
- remainder is interpreted as relative to the detected runtime prefix.
+ After a lengthy discussion, and an even lengthier time to mull over the
+ problem and its best solution, and then more discussions, we eventually
+ decided to introduce support for the magic sequence %(prefix)/. If a
+ path starts with this, the remainder is interpreted as relative to the
+ detected (runtime) prefix. If built without runtime prefix support, Git
+ will simply interpolate the compiled-in prefix.

- This solves the problem, but what new problems does it stir up? Here are
- the two most obvious ones:
-
- - What if Git was not compiled with support for a runtime prefix?
-
- In that case, we will simply use the compiled-in hard-coded prefix.
-
- - What if a user _wants_ to specify a path starting with the magic
- sequence?
-
- In that case, the user will simply need to prefix the magic sequence
- with ./ and voilà, the path won't be expanded.
+ If a user _wants_ to specify a path starting with the magic sequence,
+ they can prefix the magic sequence with ./ and voilà, the path won't
+ be expanded.

@@ Documentation/config.txt: pathname::
is expanded to the value of $HOME, and ~user/ to the specified user's home directory.
++
-+If a path starts with <RUNTIME-PREFIX>/, the remainder is -+interpreted as a path relative to Git's "runtime prefix", i.e. relative -+to the location where Git itself was installed. For example, -+<RUNTIME-PREFIX>/bin/ refers to the directory in which the Git -+executable itself lives. If Git was compiled without runtime prefix -+support, the compiled-in prefix will be subsituted instead. In the -+unlikely event that a literal path needs to be specified that should -+_not_ be expanded, it needs to be prefixed by ./, like so:
-+./<RUNTIME-PREFIX>/bin.
++If a path starts with %(prefix)/, the remainder is interpreted as a ++path relative to Git's "runtime prefix", i.e. relative to the location ++where Git itself was installed. For example, %(prefix)/bin/ refers to ++the directory in which the Git executable itself lives. If Git was ++compiled without runtime prefix support, the compiled-in prefix will be ++subsituted instead. In the unlikely event that a literal path needs to ++be specified that should _not_ be expanded, it needs to be prefixed by ++./, like so: ./%(prefix)/bin.

Variables
@@ path.c

static int get_st_mode_bits(const char *path, int *mode)
{
-@@ path.c: char *expand_user_path(const char *path, int real_home) +@@ path.c: static struct passwd *getpw_str(const char *username, size_t len)
+ * failure or if path is NULL.
+ *
+ * If real_home is true, strbuf_realpath($HOME) is used in the ~/ expansion. ++ *
++ * If the path starts with %(prefix)/, the remainder is interpreted as ++ * relative to where Git is installed, and expanded to the absolute path.
+ */
+ char *interpolate_path(const char *path, int real_home)
+ {
+@@ path.c: char *interpolate_path(const char *path, int real_home)

if (path == NULL)
goto return_null;
+
-+ if (skip_prefix(path, "<RUNTIME-PREFIX>/", &path))
++ if (skip_prefix(path, "%(prefix)/", &path))
+ return system_path(path);
+
if (path[0] == '~') {
@@ t/t0060-path-utils.sh: test_expect_success RUNTIME_PREFIX,CAN_EXEC_IN_PWD 'RUNTI
echo HERE >expect &&
+ test_cmp expect actual'
+
-+test_expect_success RUNTIME_PREFIX,CAN_EXEC_IN_PWD '<RUNTIME-PREFIX>/ works' ' ++test_expect_success RUNTIME_PREFIX,CAN_EXEC_IN_PWD '%(prefix)/ works' '
+ mkdir -p pretend/bin &&
+ cp "$GIT_EXEC_PATH"/git$X pretend/bin/ &&
-+ git config yes.path "<RUNTIME-PREFIX>/yes" &&
++ git config yes.path "%(prefix)/yes" &&
+ GIT_EXEC_PATH= ./pretend/bin/git config --path yes.path >actual &&
+ echo "$(pwd)/pretend/yes" >expect &&
test_cmp expect actual

Signature · 1 line
gitgitgadget