Why is `UpdatePackage` not idempotent / why does it compile a package even though it is already up-to-date?

[fingolfin]

In a GAP sessions with the latest IO 4.9.3 loaded:

gap> UpdatePackage("io");
#I  Retrieving PackageInfo.g from https://gap-packages.github.io/io/PackageInfo.g ...
#I  The newest version of package "io" is already installed
#I  Running compilation script on /Users/mhorn/Library/Preferences/GAP/pkg/io-4.9.3/ ...
...
#I    + gmake
#I    ./config.status gen/pkgconfig.h
#I    config.status: creating gen/pkgconfig.h
#I    echo > gen/pkgconfig.h.stamp
#I       GAC     src/io.c => gen/src/io.o
#I       GAC     gen/src/io.o => bin/aarch64-apple-darwin24-default64-kv10/io.so
true
gap>

Why does it do that?

Note that io 4.9.3 in my case is not loaded from the directory /Users/mhorn/Library/Preferences/GAP/pkg/io-4.9.3/ where PackageManager builds it, but that shouldn't matter, right? I mainly mention it in case it might be relevant for analyzing this behavior.

[fingolfin]

The same also happens for InstallPackage but there I can at least see an argument for it always re-installing the package...

What I really want is a command I can use in e.g. package CI tests to install missing dependencies, but ideally it should do nothing if the dependency is already there....

[mtorpey]

I believe this was introduced by c392c13, and was a deliberate choice.

I'm on holiday right now, so haven't reminded myself of the exact details, but I believe it was to do with IO compilation sometimes being affected by whether other packages were installed. Compiling more aggressively made things work better.

My thinking was that, if a package doesn't need to be compiled, then running the BuildPackages script (or indeed running make) should do very little work and take very little time anyway. I hope that's true?