[PATCH] Fix dependency file escaping

[Robinson, Paul]

Well, I don't think *any* of these cases come up in the wild, or gcc would have figured it out by now.  But once I had to look at this stuff anyway, I was too offended by it to leave things as they were.

GNU Make's rule is pretty straightforward, actually: backslashes escape the next character, but if a fully de-escaped filename doesn't find a match, retry with the original string.  GCC's rule is the one that's crazy, where de-escaping depends on what character follows the backslash.
--paulr

From: cfe-commits-bounces at cs.uiuc.edu [mailto:cfe-commits-bounces at cs.uiuc.edu] On Behalf Of Sean Silva
Sent: Friday, May 08, 2015 4:59 PM
To: Nico Weber
Cc: reviews+D9208+public+0ac91376ffbd3e34 at reviews.llvm.org; cfe-commits at cs.uiuc.edu
Subject: Re: [PATCH] Fix dependency file escaping

From a brief look at depfile_parser.in.cc<http://depfile_parser.in.cc>, it looks like Ninja does not correctly handle GCC's current behavior, and Paul's patch will actually make it "work". Clearly includes with `#` characters in them don't happen in practice.

Ninja's depfile_parser.cc tries to express the escaping rules as "backslash escapes certain characters", which is not the reality. Make's craziness is such that you cannot tell in isolation whether the backslashes in a substring `foo\\` need to be escaped or not without looking at the characters following that substring.

-- Sean Silva

On Fri, May 8, 2015 at 4:36 PM, Nico Weber <thakis at chromium.org<mailto:thakis at chromium.org>> wrote:
On Fri, May 8, 2015 at 4:01 PM, Paul Robinson <Paul_Robinson at playstation.sony.com<mailto:Paul_Robinson at playstation.sony.com>> wrote:
In http://reviews.llvm.org/D9208#169446, @thakis wrote:

> Does gcc intend to fix this soon? Isn't being compatible with gcc important
>  than the other things?


If gcc emitted an incorrect relocation, would you argue that it's important to be compatible with gcc?  Even if you could not point to any linker that handled that buggy relocation in a reasonable way?

'course not, but that's not the case here. If a program (say, ninja) tries to be compatible with gnu make's depfile parsing, it would previously convert "\ " to a space from what I understand. Now it's going to get "\\\ " and think that that's "\ ". So this is breaking backwards compat of clang with itself.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20150509/e47afa63/attachment.html>