Make the configuration hash the same on every machine

[adrpo]

Two machines running master at the same time both claimed the same 20 libraries and tested them twice, which is what #295 exists to prevent:

     host      | branch |  state  | count
---------------+--------+---------+-------
 ryzen-5950x-1 | master | running |    96
 ryzen-5950x-2 | master | running |    20

 libraries_claimed_by_both: 20

Not the compiler and not the library: both machines report OMCompiler v1.28.0-dev.294+gce7a5ed7bc and the same libversion (there had been no omc commit for seven hours). What differed was the confighash, which is part of the claim key (branch, libname, libversion, omcversion, confighash):

ClaRa | ryzen-5950x-1 | 1.9.0 (https://github.com/xr... | 1132175474
ClaRa | ryzen-5950x-2 | 1.9.0 (https://github.com/xr... |  247072667

Why

The hash is taken over the library's configuration, and by that point the configuration holds the reference files as an absolute path:

destinationReal = os.path.realpath(destination)
c["referenceFiles"] = destinationReal
...
confighash = strToHashInt(str(c) + hashReferenceFiles(conf["referenceFiles"]))

So two machines whose workspaces sit in different places — or whose paths resolve through different symbolic links — hash an identical configuration differently.

This was harmless for as long as it existed: each machine had its own sqlite3.db, so the hash only ever had to be stable on the machine that produced it. Sharing one database is what made it matter, and it breaks two things at once:

• the job claim does not exclude the other machine, so both test the same library — the whole point of Several improvements to LibraryTesting needed: network SQLDB and docker images to run the tests #295;
• the "we already have results for it" check misses across machines, so even a sequential run redoes work another machine has done.

The fix

The path contributes nothing the hash needs. The contents of the reference files are hashed on the very next line by hashReferenceFiles(), and the same files in a different directory are the same test. So the path is replaced by a constant before hashing.

Checked

The same configuration under two different reference directories:

before   machineA 3181697168  machineB 2732211752  -> DIFFERENT (both claim the library)
after    machineA 1884809167  machineB 1884809167  -> SAME (machines agree)
after    reference file changed          -> DIFFERENT (correct)

so the hash stops depending on where the files are while still depending on what is in them. Plus a full configs/sanityCheck.json run with the CI assertion passing.

One deliberate consequence

Every confighash changes value once. The first run after this sees no matching rows and treats every library as untested, so it tests everything once. From then on the machines agree — and until then they duplicate work anyway, which is the bug.

---

Generated by Claude Code.