GHAs and precompiled headers

[Thomas Stüfe]

On Mon, Jan 24, 2022 at 10:31 AM Aleksey Shipilev <shade at redhat.com> wrote:

> On 1/24/22 10:26 AM, Thomas Stüfe wrote:
> >
> >
> > On Mon, Jan 24, 2022 at 10:15 AM Aleksey Shipilev <shade at redhat.com
> <mailto:shade at redhat.com>> wrote:
> >
> >     On 1/24/22 8:43 AM, Thomas Stüfe wrote:
> >      > We generally build without --disable-precompiled-headers in GHAs,
> which
> >      > hides errors from missing includes. Since GHAs are very useful to
> test
> >      > builds on side platforms, would it not make sense to build without
> >      > precompiled headers? Or is that too costly?
> >
> >     But... We *do* build Hotspot without PCH in "additional"
> configurations that test different
> >     platforms:
> >
> >
> https://github.com/openjdk/jdk/blob/9bf6ffa19f1ea9efcadb3396d921305c9ec0b1d1/.github/workflows/submit.yml#L408-L432
> >     <
> https://github.com/openjdk/jdk/blob/9bf6ffa19f1ea9efcadb3396d921305c9ec0b1d1/.github/workflows/submit.yml#L408-L432
> >
> >
> > Yes, but only Linux. Leaves out macOS and Windows. For people who mainly
> use Linux and rely on GHA
>
> Ah.
>
> The last I checked, dropping PCH on Windows made build times suffer a lot.
> Given how underpowered
> are the Windows and MacOS build nodes in GHA, I'd rather accept a small
> possibility of non-PCH
> breakage than slowness of generic GHA workflows.
>
>
Pity, but I guess it cannot be avoided. It hit me this weekend, but at
least I know now the GHA limitations.

Thanks, Thomas