fix: Correctly create directories for paths with forward slashes on Windows

[Mixaill]

Fixes sentry-native initialization on Windows when databasepath has forward slashes.

[Swatinem]

Nice catch! We should have a unit test for this.

[Mixaill]

@Swatinem updated

[Swatinem]

Having this tested would be nice. I think its enough to just add a /qux or something here:

sentry-native/tests/unit/test_path.c

Line 10 in 5020ff5

sentry_path_t *nested2 = sentry__path_join_str(nested, "baz");

[Mixaill]

@Swatinem tests added.

I do not known, if backslashes should be tested on non-windows platforms too?

update: it works.

[Swatinem]

Sorry for the back and forth here…

Unices are apparently very forgiving when it comes to paths, but you do have an error in there.

/home/vsts/work/1/s/tests/unit/test_path.c:150:46: warning: incompatible pointer types passing 'int [4]' to parameter of type 'sentry_pathchar_t *' (aka 'char *') [-Wincompatible-pointer-types]
    sentry_path_t *path_1 = sentry__path_new(L"foo");
                                             ^~~~~~
/home/vsts/work/1/s/src/sentry_path.h:220:37: note: passing argument to parameter 's' here
sentry__path_new(sentry_pathchar_t *s)
                                    ^

sentry__path_new uses a different typedef for the chars per platform. I think a better test is to compare the behavior of sentry__path_from_str with calling sentry__path_join_str directly, and see that they match up.

[Mixaill]

sentry__path_new uses a different typedef for the chars per platform

fixed

I think a better test is to compare the behavior of sentry__path_from_str with calling sentry__path_join_str directly, and see that they match up.

My point that sentry_path_windows.c::sentry__path_create_dir_all() does not call these functions at all. So it maybe worth to add such kind of test, but it is a bit out of scope for this PR.

[Swatinem]

The reason I would like to compare to manually calling sentry__path_join_str is that I think the testcase is buggy on unix right now.

sentry-native/src/path/sentry_path_unix.c

Line 322 in aca8890

if (*ptr == '/' && ptr != p) {

I think unixes will just happily create a directory named "foo\bar", which is different from "foo" / "bar".

[Swatinem]

Which reminds me… should we maybe normalize separators on unix as well? How common is it to use \ as separator on unix? Maybe @mitsuhiko can comment here, since you know a lot about path encodings.

[Mixaill]

I think the testcase is buggy on unix right now.

Thanks, now I understood it. Yeah, it is definitely broken, because \ is allowed character on unix.

How common is it to use \ as separator on unix?

At least it's possible that someone actually uses the backslash as a folder name in unix --> it will break sentry-native for him.

The reason I would like to compare to manually calling sentry__path_join_str

It will be failed, because sentry__path_from_str does not perform normalization. Maybe we should add it at least for Windows?

[mitsuhiko]

We should only execute tests with \\ as separators on windows and skip this on Linux. We should not convert backslashes to forward slashes on UNIX. That's unexpected and may cause issues if someone actually wants to use them.

[Mixaill]

We should only execute tests with \ as separators on windows and skip this on Linux.

Done