[cfe-commits] r159256 - in /cfe/trunk: include/clang/Basic/FileManager.h lib/Basic/FileManager.cpp
Axel Naumann
Axel.Naumann at cern.ch
Thu Jun 28 03:04:02 PDT 2012
Hi Matthieu,
On 06/27/2012 07:18 PM, Matthieu Monrocq wrote:
> On Wed, Jun 27, 2012 at 4:15 PM, Axel Naumann <Axel.Naumann at cern.ch
> <mailto:Axel.Naumann at cern.ch>> wrote:
> Okay. I see three maybe more reasonable options, in decreasing order of
> preference for us:
> * truncate the cache (we'd anyway only remove the last N files)
> * add interface to decide which file should be cached and which
> shouldn't, based on the file's full name, in our case we'd likely only
> cache read-only files (e.g. STL).
> * clear the cache altogether. That's the least performant option for us.
>
> Should we implement the first option? Or keep the current version?
>
> Cheers, Axel.
> _______________________________________________
> cfe-commits mailing list
> cfe-commits at cs.uiuc.edu <mailto:cfe-commits at cs.uiuc.edu>
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
>
>
> There has been a discussion about avoiding warnings in 3rd party headers
> on top of STL headers.
Sorry, I don't see yet how that's relevant here.
> In this case we could expect that Cling did the
> smart things and cached 3rd party headers but not local ones. Or perhaps
> had the ability to only clear ones "own" files.
Isn't that what I was suggesting as option 2? Are you saying that I
missed an existing interface that enables selective caching?
> From what I understood, in Cling you don't edit many files at once.
> However if we are talking about being able to react to a modification to
> a header file, without really knowing its dependencies, then the
> truncation cannot be used easily here it seems.
We would use truncation as a simplistic dependency analysis: any file
that comes after might depend on the file that needs to be removed. The
current trunk enables removal of a specific FileEntry. With that we
could do more fine-grained analysis, a la include-what-you-use. And
that's really our preferred choice.
Cheers, Axel.
More information about the cfe-commits
mailing list