[PATCH] PR21482: long paths on Windows

[Robinson, Paul]

> -----Original Message-----
> From: Rafael Espíndola [mailto:rafael.espindola at gmail.com]
> Sent: Tuesday, November 11, 2014 10:09 AM
> To: Robinson, Paul
> Cc: Michael Spencer; Aaron Ballman; llvm-commits at cs.uiuc.edu
> Subject: Re: [PATCH] PR21482: long paths on Windows
> 
> Why is the check on the absolute path being larger than the limit?
> 
> If the current directory path is large, an operation using a short
> relative path will still work, no? 

No. I tried a 243-character relative path and made it not convert to an
absolute path, when current directory had 36 characters. This failed 
with ERROR_PATH_NOT_FOUND (becomes errc::no_such_file_or_directory).
However the same path, if converted to \\?\ absolute, succeeds.

It's entirely possible that I've screwed up the test somehow, but I
tried this a couple times with different combinations and got the
same outcomes.

So, what matters is the length of the absolute path of the entity
being created, not the length of the string being passed to the API.
--paulr

> The problem, if I understand it
> correctly, happens when the path (relative or not ) that will be
> handled to a windows API is too long.
> 
> On 11 November 2014 10:27, Robinson, Paul
> <Paul_Robinson at playstation.sony.com> wrote:
> > New patch attached.
> >
> > I ended up not using make_absolute because it caused more string
> copying.
> > If there's some reason to consider it safer than current_path and doing
> > my own pasting, I can do that instead, but only if y'all think it has
> > actual value.
> >
> > This changed the specific error returned by another fs test, which
> > seems reasonable.
> > --paulr
> >
> >> -----Original Message-----
> >> From: Michael Spencer [mailto:bigcheesegs at gmail.com]
> >> Sent: Monday, November 10, 2014 4:04 PM
> >> To: Robinson, Paul
> >> Cc: Aaron Ballman; Rafael Espíndola; llvm-commits at cs.uiuc.edu
> >> Subject: Re: [PATCH] PR21482: long paths on Windows
> >>
> >> On Mon, Nov 10, 2014 at 3:05 PM, Robinson, Paul
> >> <Paul_Robinson at playstation.sony.com> wrote:
> >> >> -----Original Message-----
> >> >> From: Aaron Ballman [mailto:aaron.ballman at gmail.com]
> >> >> Sent: Monday, November 10, 2014 2:47 PM
> >> >> To: Rafael Espíndola
> >> >> Cc: Robinson, Paul; llvm-commits at cs.uiuc.edu
> >> >> Subject: Re: [PATCH] PR21482: long paths on Windows
> >> >>
> >> >> On Mon, Nov 10, 2014 at 5:43 PM, Rafael Espíndola
> >> >> <rafael.espindola at gmail.com> wrote:
> >> >> > static std::error_code widen_path(const Twine &Path8,
> >> >> >
> >> >> > widenPath
> >> >> >
> >> >> > Using  '\\?\'  also disables using / as a path separator, no?
> Don't
> >> we
> >> >> > have to check/assert that there is no / in Path8?
> >> >
> >> > So for paranoia I should translate all / to \ in the widened path?  I
> >> don't
> >> > see any existing utility to do that, but easy enough to invent one.
> >>
> >> sys::path::native(). in Path.h
> >>
> >> We also have make_absolute in FileSystem.h
> >>
> >> - Michael Spencer
> >>
> >> >
> >> >> >
> >> >> > +  size_t NLevels = ((248 - TmpLen) / 10) + 1;
> >> >> > +  const char *OneDir = "\\123456789";
> >> >> >
> >> >> > Use the size of OneDir instead of hard codding the 10.
> >> >> >
> >> >> > Aaron, this only uses '\\?\' to create long paths. They can still
> be
> >> >> > accessed by regular tools by changing the current directory, no?
> >> >>
> >> >> Yes, relative paths aren't an issue. Just tools using absolute paths
> >> >> that don't support the extended path namespace (which can happen if
> >> >> the app assumes MAX_PATH is actually the max).
> >> >>
> >> >> ~Aaron
> >> >>
> >> >> >
> >> >> > On 10 November 2014 16:01, Robinson, Paul
> >> >> > <Paul_Robinson at playstation.sony.com> wrote:
> >> >> >> Support directory names longer than 248 characters on Windows.
> >> >> >>
> >> >> >> The normal Windows path limit is 260 characters; the limit for
> >> >> directory
> >> >> >> names is 248, to leave room for an 8.3 filename at the end.
> Adding
> >> the
> >> >> >> '\\?\' prefix greatly expands these limits.  Intercept path names
> >> that
> >> >> >> we are about to pass to the Windows APIs and add the prefix if
> >> >> necessary.
> >> >> >>
> >> >> >> Also correct the comment about the state of the temporary
> directory
> >> >> used
> >> >> >> by the Support unittests.
> >> >> >> --paulr
> >> >> >>
> >> >> >>
> >> >> >> _______________________________________________
> >> >> >> llvm-commits mailing list
> >> >> >> llvm-commits at cs.uiuc.edu
> >> >> >> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
> >> >> >>
> >> >
> >> > _______________________________________________
> >> > llvm-commits mailing list
> >> > llvm-commits at cs.uiuc.edu
> >> > http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits