RFC: Support for preferring paths with forward slashes on Windows

Project Infrastructure LLVM Dev List Archives
1

Hi,

When using Clang on Windows as a drop-in replacement for GCC, one issue that crops up fairly soon is that not all callers can tolerate paths spelled out with backslashes.

This is an issue when e.g. libtool parses the output of "$CC -v" (where clang passes an absolute path to compiler-rt libraries) and uses parts of that in shell script contexts that don't tolerate backslashes, when some callers call "$CC --print-search-dirs", etc.

This is also one of the most important things that MSYS2 patches in their distribution of Clang/LLVM according to their patch tracker [1].

(I've locally worked around this in my distribution without patching, by filtering clang's stdout in a wrapper, when options like "-v" or "--print-search-dirs" are detected, but that's essentially the same as patching.)

I've finally taken the plunge and tried to implement this properly. I've got a decent patch set [2] that I could start sending for review, but before doing that, I'd like to discuss the overall design.

The main idea is that I add a third alternative to path::Style - in addition to the existing Windows and Posix path styles, I'm adding Windows_forward, which otherwise parses and handles Windows paths like before (i.e. accepting and interpreting both separators), but with a different preferred separator (as returned by get_separator()).

This allows any code on any platform to handle paths in all three forms, just like in the existing design, when explicitly giving a path::Style argument.

To actually make it have effect, one can make path::Style::native act like Windows_forward instead of plain Windows. I'm not entirely sure what the best strategy is for when to do that - one could do it when LLVM itself was built for a MinGW target (which kind of breaks the assumption that the tools work pretty much the same as long as one passes the right --target options etc), or one could maybe set it up as a configure time CMake option? Or even make it a globally settable option in the process, to allow changing it e.g. depending on the tool's target configuration?

I also faintly remember that Reid at some point implied that it could be an option to switch all Windows builds outright to such a behaviour?

Most of the code is entirely independent of the policy decision of when/where to enable the behaviour - the decision is centralised to one single spot in LLVMSupport.

In any case, with this design and a quite moderate amount of fixups, most of the tests in check-all seem to pass, if switching the preference.

There's a couple tests that fail due to checking e.g. the literal paths %s or %t (as output by llvm-lit, with backslashes) against paths that the tools output. There's also a dozen or so of tests in Clang (mainly regarding PCH) that seem to misbehave when the same paths are referred to with varying kinds of slashes, e.g. stored with a forward slash in the PCH but referred to with backslashes in arguments to Clang, where paths are essentially equal but the strings differ. (For actual use with PCH, Clang built this way seems to work - and MSYS2 have been running with tools patched this way for quite some time, and I haven't heard about reports about bugs relating to that patch.)

If the design seems sane (have a look at [2] if you want to have a look at my whole series at the moment) I'd start sending the initial patches for review.

// Martin

[1] MINGW-packages/README-patches.md at master · msys2/MINGW-packages · GitHub

[2]Â https://github.com/llvm/llvm-project/compare/main...mstorsjo:path-separator

2

I could be mistaken, but I believe that since the dawn of time, Windows has just secretly supported forward slashes. A quick google search does not turn up any Microsoft docs stating that this is true, but I've heard rumors that it's been this way since DOS. On my Windows 10 machine, Powershell accepts /, cmd.exe accepts /, and Visual Studio accepts /. Whomever takes it upon themselves to work on this should test extensively before committing code. I would probably feel better if somebody could dig up some authoritative source on this.

Assuming that this is the case, it would probably be nice if any paths we take in were just immediately canonicalized to use / and all paths just have forward slash. I know we have a ton of tests that have this `{(/|\\)}` regex in them, and it would be nice if we could just not do that.

Thanks,
  Chris Tetreault

3

The Win32 userspace (usually) supports them, but the underlying NT kernel does not. So they are normally converted, but not always. This can affect a few places, such as paths whose names begin with the character sequence \\?\ or, if I recall correctly, a few odd places such as LoadLibrary(Ex) not supporting paths containing / (explicitly documented at https://docs.microsoft.com/en-us/windows/win32/api/libloaderapi/nf-libloaderapi-loadlibrarya)

4

I could be mistaken, but I believe that since the dawn of time, Windows has just secretly supported forward slashes. A quick google search does not turn up any Microsoft docs stating that this is true, but I've heard rumors that it's been this way since DOS. On my Windows 10 machine, Powershell accepts /, cmd.exe accepts /, and Visual Studio accepts /.

Yes, overall most APIs that take paths can take either form, but in my experience, cmd.exe pretty exclusively requires backslashes.

Whomever takes it upon themselves to work on this should test extensively before committing code. I would probably feel better if somebody could dig up some authoritative source on this.

I don't think this aspect is anything new/controversial wrt LLVM so far; it can take paths that use forward slashes (if given such paths) in a number of places and pass them through pretty much as-is to the underlying APIs.

But most cases where we take a path and feed to a underlying API is well centralised to a single function, which takes our char based UTF8 paths and widens them to UTF16 wchar_t, before passing them to the actual Win32 APIs in that form. Currently, that function forces the paths to backslash form in certain cases (when it needs to prepend a \\?\ prefix for long paths), but if we felt weary about it we could make it always force them to backslash form.

So the fact that we can pass paths with forward slashes to Win32 APIs is a preexisting condition and nothing that my patch set would change, essentially - it'd just do it more often than before.

Assuming that this is the case, it would probably be nice if any paths we take in were just immediately canonicalized to use / and all paths just have forward slash. I know we have a ton of tests that have this `{(/|\\)}` regex in them, and it would be nice if we could just not do that.

If desired, that could be a later goal - that's a couple steps further than what I aimed for so far though.

Right now, my patchset canonicalizes paths that are made up internally (functions like current_path(), getMainExecutable(), findProgramByName(), and how InitLLVM() sets argv[0]) and uses the preferred separator wherever paths are assembled in code, but in many cases, paths are taken in and passed around in the user-provided form too. Given the full interface of e.g. Clang, there's a huge number of different places where paths can be provided (there's dozens of various command line options that take paths as arguments).

Also, judging from both GCC and MSVC, neither of them seem to canonicalize paths on input. If I call either of them with e.g. c:\dir\source.c or c:/dir/source.c, then the warnings emitted from that file are printed with slashes in the exact form I input.

But in any case, regardless of how far we want to go with canonicalizations in either form, the patchset I've started on, given that others agree on the design, is a first step towards being able to use forward slashes. It works quite well to apply it gradually until switching the preference.

// Martin

5

Big thanks to you for making investigation in this area!

clang/test/Driver tests suffer the most from Windows backslashes. MC and
DebugInfo suffer a bit as well.
I have seen so many times a new test did not pass on Windows and a fixup follow-up was needed.
Sometimes the author may adjust the test and slighly degrade the test
quality if they cannot figure out the best way supporting both / and \
(using {{/|\\\\}} multiple times on one line could clutter up).

6

Thanks for working on this.

I also noticed is that the paths printed by clang -v or -### are
escaping the backslashes and put them into quotes, i.e.
"C:\\path\\to\\clang.exe" -cc1 "..\\special'^`character .c"
Interestingly, it still works copy&pasting it to the Windows command
line [2], but cmd.exe's escape character is ^ and PowerShell's is the
backtick `. What would the correct output be?

[1] batch file - Documented behavior for multiple backslashes in Windows paths - Stack Overflow

Michael

7

I wasn't aware that cmd.exe had an escape char per se (other than ^ for line continuations?).

The fact that such slashes are printed double is, iirc, an intentional quirk, so that the command lines are copypasteable in a variety of contexts: cmd.exe don't need them doubled (but tolerates them), bash unescapes them so it also can execute them.

This is actually one downside of using forward slashes, as cmd.exe wouldn't be able to execute such a command straight out (only the slash direction of the command executable itself matters though).

// Martin

8

Thanks for working on this! I think this mode is a long time coming.

I also noticed is that the paths printed by clang -v or -### are
escaping the backslashes and put them into quotes, i.e.
“C:\path\to\clang.exe” -cc1 "…\special’^character .c" Interestingly, it still works copy&pasting it to the Windows command line [2], but cmd.exe's escape character is ^ and PowerShell's is the backtick . What would the correct output be?

I wasn’t aware that cmd.exe had an escape char per se (other than ^ for
line continuations?).

The fact that such slashes are printed double is, iirc, an intentional
quirk, so that the command lines are copypasteable in a variety of
contexts: cmd.exe don’t need them doubled (but tolerates them), bash
unescapes them so it also can execute them.

Yep, I was going to say that.

This is actually one downside of using forward slashes, as cmd.exe
wouldn’t be able to execute such a command straight out (only the slash
direction of the command executable itself matters though).

I think as long as the user has a way to choose between the styles, they’ve got what they need to unblock themselves.

9

I wasn't aware that cmd.exe had an escape char per se (other than ^ for
line continuations?).

Some MS docs recommend to use it to escape redirection (<, >) and pipe

symbols [1]. Interestingly, ^ does NOT escape within double-quotes.

The best explanation I found about it is [2].

[1] echo | Microsoft Docs
[2] Quotes, Escape Characters, Delimiters - Windows CMD - SS64.com

Michael

10

Just FTR, as there weren't any specific comments on the implementation strategy, I went ahead and posted the initial couple patches for review, at ⚙ D111879 [Support] Add a new path style for Windows with forward slashes and ⚙ D111880 [Support] [Windows] Convert paths to the preferred form.

// Martin