[NFC][SYCL] Move visitors into their own class. #2081
Conversation
A recent patch has shown there is good reason to have access to Sema/ASTContext when doing the visiting. This also makes it more consistent with the other visitors in Clang. Additionally, there was an odd bracket issue that I've fixed, seemingly a close/open namespace got lost along the way, so the 'close' namespace (with comment) bracket was ACTUALLY the end of the function.
erichkeane
requested review from
elizabethandrews,
Fznamznon and
premanandrao
as code owners
Interestingly, the clang-format on my local machine doesn't suggest this, and apparently didn't when this code was written initially. THe comment layout here is SUPER weird, I'm not a huge fan of it, but if clang format says it, *shrugs*.
Yep, it will. In order to properly canonicalize an array type you need access to ASTContext (in SemaRef). I presume we'll also see other uses for Sema in the visitors in the future. |
|
FYI. CI check uses clang-format-9, which seems to be the most recent version available on Ubuntu 18.04 from standard repo. |
Mine was apparently quite old (3.8 release!), so perhaps that is it. I've updated mine to 11, so hopefully I won't have this problem again :) |
|
AFAIK, there were some inconsistencies in formatting between the version built from |
This expands our CI to test the loader; the dispatcher that is used when multiple adapters are availabe. The old "run on hardware" jobs should behave the same (they have the loader tests disabled), but there are new "combined" jobs with OpenCL/Level Zero + Native CPU. Closes: #2081
A recent patch has shown there is good reason to have access to
Sema/ASTContext when doing the visiting. This also makes it more
consistent with the other visitors in Clang.
Additionally, there was an odd bracket issue that I've fixed, seemingly
a close/open namespace got lost along the way, so the 'close' namespace
(with comment) bracket was ACTUALLY the end of the function.