Go version

6f44cc8

Output of go env in your module/workspace:

N/A

What did you do?

See #65545.

What did you see happen?

https://go.dev/cl/559615 introduced a regression on windows, but the CL had a TryBots-Pass+1 vote.

I examined the Checks tab to see which builders had run, and saw:

• tricium
• x_pkgsite-go1.20-linux-amd64
• x_pkgsite-go1.21-linux-amd64
• x_pkgsite-go1.22-linux-amd64
• x_pkgsite-gotip-linux-amd64
• x_pkgsite-gotip-linux-amd64-boringcrypto
• x_pkgsite-gotip-linux-amd64-misccompile
• x_pkgsite-gotip-linux-amd64-misccompile-test_only
• x_pkgsite-gotip-linux-amd64-race
• x_pkgsite-gotip-windows-amd64

What did you expect to see?

Since x/pkgsite/cmd/pkgsite is intended to be usable as a replacement for x/tools/cmd/godoc (#59347, #40371, #59056), it should have pre-commit testing on at least all of the first class ports that have scalable builders. Namely:

• linux/386
• linux/amd64
• linux/arm
• linux/arm64
• windows/386
• windows/amd64

In addition, it is surprising that the release-branch and -race variants were only run for linux, since all of those variants also exist for windows too.

Adding the release-branch builders, 386 architecture, and -race configuration would have increased the chances of catching the regression by a factor of 6.