Windows binary build: mtime on source-files is newer than base.cache

[timholy]

Formerly Revise.track(Base) was used only by people hacking on Julia. Now with Rebugger, it's used behind the scenes by any user who steps into a method defined in Base. So the performance of Revise.track(Base) matters much more now than it used to. For some of us it's pretty quick (a fraction of a second), but it turns out to be glacially slow for others (>10s). In timholy/Revise.jl#168 we tracked it down to a difference of (1) people who build from source vs people who download binaries AND (2) Linux vs Windows.

While details are provided in that linked issue, a brief summary is this: Revise detects whether a file has been edited based on an mtime criterion, comparing against the value written into the base.cache file (see #24120). On Linux binaries, the mtime difference seems to be 0; however, on Windows it seems that many of the source files are 4-10 seconds newer than the time stamp in base.cache.

Is something that's easily fixable? Could we just change the order of a couple of steps in the build so that the source files get copied before the base.cache file is generated?

[nalimilan]

AFAICT we can't really install files before building the cache as the former happens during make install, which depends on make release. But maybe we just need to adjust the copy command to preserve timestamps (equivalent to cp -a on Unix)?