[cfe-dev] FW: Patch for UCNs in string literals

[AlisdairM(public)]

Oops, intended for list, not private email!

AlisdairM

-----Original Message-----
From: AlisdairM(public) [mailto:public at alisdairm.net] 
Sent: 05 July 2009 10:48
To: 'Eli Friedman'
Subject: RE: [cfe-dev] Patch for UCNs in string literals

> -----Original Message-----
> From: cfe-dev-bounces at cs.uiuc.edu [mailto:cfe-dev-bounces at cs.uiuc.edu]
> On Behalf Of Eli Friedman
> Sent: 05 July 2009 07:58
> To: James Porter
> Cc: cfe-dev at cs.uiuc.edu
> Subject: Re: [cfe-dev] Patch for UCNs in string literals
> 
> On Sat, Jul 4, 2009 at 10:16 PM, James Porter<porterj at alum.rit.edu>
> wrote:
> > Eli Friedman wrote:
> >> Please just include the subset of tests that currently passes; we
> can
> >> add additional tests as the issues get fixed.
> >
> > I was under the impression based on the website that XFAIL'd tests
> were
> > encouraged:
> >
> >     http://clang.llvm.org/OpenProjects.html
> >
> >     "Pick a paragraph, write a few tests, and see if they work! Even
> if
> >     they don't we'd still like the new tests (with XFAIL'd) so that
> we
> >     know what to fix."
> 
> A patch should be accompanied by a non-XFAIL'd test for each thing it
> fixes; otherwise, the tests won't detect regressions.  I haven't
> thought through what that means for the C++ tests; perhaps splitting
> them into p2a/p2b/p2c/etc. would be appropriate.

Well, it looks like precedent is upon us!

I'd appreciate some guidance on how best to submit these tests.  There are at least two issues here:

i/ there is a different behaviour between C++ and C.  Ideally I would like to test both.
That is why I submitted two versions of tests for the same paragraph.

In hindsight it looks like we might be better served by a parallel test structure to test/CXX for the C language, although I don't have a copy of that standard myself to know where to submit tests.  At least with C++0x we have publicly available working draughts.

A subset of this issue is the change between C++ and C++0x, which again suggests two tests.  I followed precedent elsewhere of tagging the 0x specific tests with the cpp0x suffix.  Ideally we should run all C++ tests in both C++ and C++0x mode as (most) conforming C++03 programs should also be conforming C++0x programs.  So far, the main incompatibility would be handling of auto.

ii/ Eli's original issue that we want to see tests passing to indicate the new feature works, as well as the extensive tests to validate the paragraph.  I am happy to break out the passing functionality tests, but I don't want to lose the full paragraph coverage.  How should we name these tests and where should we place them?

For example, when working on my type-alias patch I am using a test file unrelated to paragraph numbers, as the controlling paragraph says not a lot more than 'works like a typedef' so I need to test a wide variety of typedef-like uses, which is almost an open set.  Could I submit my working tests as n2170a.cpp, while retaining the full XFAILed p2?  Likewise, we can incrementally add n2170b,c,d,etc. as support for UCNs in more contexts becomes available.

AlisdairM