MSVC bots: Chromium, PDFium, and V8

[K. Moon]

Hi infra folks,

Chromium doesn't support MSVC anymore, but PDFium and V8 explicitly do, and take some dependencies from Chromium, like //third_party/zlib.

I recently ran into an issue with MSVC compilation while rolling DEPS for PDFium (see crrev.com/c/2161475 for more details). The breaking change to //third_party/zlib wasn't caught by the MSVC bots. (Some speculation as to why on https://pdfium-review.googlesource.com/c/pdfium/+/69002.) This required a fix to third_party/zlib/BUILD.gn before the PDFium roll succeeded. V8 likely would have had the same issue when trying to roll zlib.

As long as PDFium and V8 continue to maintain MSVC support, I was wondering if there's some way we could make sure the Chromium bots catch such things? Right now, they don't seem to catch changes to zlib.

--Kahmy

[Takuto Ikuta]

If you see such issues frequently, I think you can file an issue to set up the bot to detect that.

But I currently think fixing after CQ failure in your repository seems reasonable.

+Nico Weber 

Do you have any thoughts for this?

--
You received this message because you are subscribed to the Google Groups "infra-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email to infra-dev+...@chromium.org.
To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/infra-dev/CACwGi-5gsTqbHB4WnPHWHTb_7LKkemwwfzBXe5KUFvG95GRkiA%40mail.gmail.com.

[Lei Zhang]

We recently hit 2 MSVC issues in zlib. Aside from that, most of the prior PDFium MSVC issues have been with V8. IIRC, V8's MVSC bot builds 64-bit release, whereas PDFium bots build 32-bit and 64-bit, in debug mode, so they do not catch the same set of issues.

Here are all the V8 MSVC fixes from the last 2 years for issues we found by rolling V8 in PDFium:

https://chromium-review.googlesource.com/c/v8/v8/+/1964391
https://chromium-review.googlesource.com/c/v8/v8/+/1960906
https://chromium-review.googlesource.com/c/v8/v8/+/1960905
https://chromium-review.googlesource.com/c/v8/v8/+/1899547
https://chromium-review.googlesource.com/c/v8/v8/+/1697043
https://chromium-review.googlesource.com/c/v8/v8/+/1329694
https://chromium-review.googlesource.com/c/v8/v8/+/1187702

[Nico Weber]

My thought is that we should explicitly not support this in Chromium. We've talked before about how if you want chroimum's zlib to be some general zlib fork, then src.git isn't the ideal place to host it. It also isn't the worst place, but it comes with tradeoffs, and this is one of them.

IMHO clang-cl is very stable by now, and MSVC ships with clang-cl as of MSVC 2019 so we should reevaluate if we need to support msvc in pdfium, v8, etc.

[Adenilson Cavalcanti]

As they are still available through manual selection, I would like to
ask what is the purpose of the MSVC bots in Chromium's infrastructure?

It seems they don't actually build the code and execute the steps up
to 'analyze':
https://ci.chromium.org/p/webrtc/builders/try/win_x64_msvc_rel/19172?

> To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/infra-dev/CAMGbLiEofzvqdmoDpo0uC%3DwJrLUQdNVKu425f3Acxsb_LWa8Ug%40mail.gmail.com.

[Nico Weber]

They're there to check that the !is_clang bits of the _gn_ build files mostly work, since those bits are used by non-chromium projects.

[K. Moon]

Sounds like other than maybe checking again if third parties using PDFium/V8 can switch to using clang-cl, the conclusion is to stick with the status quo.

[Dirk Pranke]

+1 to what thakis@ wrote. 

-- Dirk

To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/infra-dev/CACwGi-525cTt5NR-x1weA8rhTi9-6dJPtt5YsOgh_JW%3DN8MpMA%40mail.gmail.com.