Subject: Re: [boost] ... broken with clang
From: Mathias Gaunard (mathias.gaunard_at_[hidden])
Date: 2011-12-17 19:25:14
On 12/18/2011 01:12 AM, Michael Caisse wrote:
> While I have no intent to ship product from clang anytime soon, I do
> find that its less verbose and more targeted error messages can provide
> a quick way to sort through the gcc/msvc cruft from template errors.
Have you really tried it?
Because I have the exact opposite experience.
When templates are involved, it generates much more verbose output with
a huge amount of cruft (especially when both macros and templates are
involved, as is often the case in Boost). While with GCC the error
message is in the middle of the screen, with Clang you have to scroll
two or three pages.
It's also much harder to make it display types or other compile-time
values in error messages or warnings.
> If one were to look back through the ML archives when Doug was getting
> boost to build with clang you will find that many a non-compliant
> construct was found via clang.
Clang is indeed a good way to find invalid code. Its standard
conformance is very good while still being very GCC-compatible.