Some temporary directory handling changes #94
Conversation
clar/sandbox.h
Outdated
| @@ -2,7 +2,8 @@ | |||
| #include <sys/syslimits.h> | |||
| #endif | |||
|
|
|||
| static char _clar_path[4096 + 1]; | |||
| #define CLAR_PATH_MAX 4096 | |||
| static char _clar_path[CLAR_PATH_MAX]; | |||
There was a problem hiding this comment.
Is it intentional that we lose the 4097th byte here?
| #else | ||
| char tmp[CLAR_PATH_MAX]; | ||
|
|
||
| if (realpath(buffer, tmp) == NULL) |
There was a problem hiding this comment.
I worry a bit that realpath may not be available on all systems, as it is only specified by POSIX.1-2001 and thus not part of C90 itself. And unfortunately we have folks that compile Git on systems that predate POSIX.1-2001 itself.
I sometimes wish that some kind of resource existed that provides a matrix of operating system <-> availability of certain features.
clar/sandbox.h
Outdated
| #define CLAR_PATH_MAX 4096 | ||
| static char _clar_path[CLAR_PATH_MAX]; | ||
| static char _clar_tempdir[CLAR_PATH_MAX]; | ||
| size_t _clar_tempdir_len; |
There was a problem hiding this comment.
Does this need to be non-static? I couldn't spot any uses of it outside of the current scope.
clar/sandbox.h
Outdated
| return 0; | ||
| } | ||
|
|
||
| static int clar_sandbox_create(void) | ||
| { | ||
| char alpha[] = "0123456789abcdef"; |
clar/sandbox.h
Outdated
| _clar_sandbox[_clar_tempdir_len + 6] = alpha[(num & 0x00000f00) >> 8]; | ||
| _clar_sandbox[_clar_tempdir_len + 7] = alpha[(num & 0x000000f0) >> 4]; | ||
| _clar_sandbox[_clar_tempdir_len + 8] = alpha[(num & 0x0000000f) >> 0]; | ||
| _clar_sandbox[_clar_tempdir_len + 9] = '\0'; |
There was a problem hiding this comment.
Now that we have a central directory where all test results will be written into, do we need randomness at all? We could compute statically derived paths, e.g. by appending the test name to the sandbox path. These have to be unique anyway, so the only thing that we need to care about is to prune such a directory that may be left over from a previous run.
There was a problem hiding this comment.
This is a really excellent suggestion, which I have implemented. Test names are much more readable than random digits. 🙏
ethomson
force-pushed
the
ethomson/tempdirs
branch
2 times, most recently
from
0d03c04 to
97a3772
Compare
ethomson
force-pushed
the
ethomson/tempdirs
branch
2 times, most recently
from
4b50259 to
0761710
Compare
|
I refreshed this because in libgit2, we tried to update to the latest clar which failed, because we rely on understanding what the sandbox dir is on Windows. Since there's concern about the |
ethomson
force-pushed
the
ethomson/tempdirs
branch
from
e15b8af to
d9cd8ca
Compare
We previously canonicalized the temporary directory on macOS (since `/tmp` is always `/private/tmp`). Expand this to Windows and - optionally - anybody who has `realpath`. This ensures that tests who want to understand where their sandbox is can accurately do so.
clar now has the concept of a _temporary directory_, which is the temporary directory for the entirety of a `clar` invocation (a set of test runs) and a sandbox, which is the temporary directory for a single test invocation. This allows us to ensure that a well-written test (that only writes into its sandbox) doesn't poison the well for future test invocations if it fails to clean up its sandbox.
Move out of the sandbox directory before trying to remove it; this is required on Windows, where you cannot remove a directory that you're chdir'd into.
ethomson
force-pushed
the
ethomson/tempdirs
branch
from
d9cd8ca to
a17a427
Compare
Some changes to temporary directory handling — general quality-of-life changes with an eye toward eventually supporting parallel execution (presumably with separate tempdirs per thread).
canonicalize temp sandbox directory everywhere
We currently only canonicalize the temp sandbox directory on macOS,
which is critical since
/tmpis really/private/tmp. However, weshould do it everywhere, so that tests can actually expect a consistent
outcome by looking at
clar_sandbox_path().tempdir and sandboxes are now separate entities
clar now has the concept of a temporary directory, which is the
temporary directory for the entirety of a
clarinvocation (a set oftest runs) and a sandbox, which is the temporary directory for a single
test invocation.
This allows us to ensure that a well-written test (that only writes into
its sandbox) doesn't poison the well for future test invocations if it
fails to clean up its sandbox.