During garbage collection we cache several things -- a set of known-dead
paths, a set of known-alive paths, and a map of paths to their derivers.
Currently they use STL maps and sets, which are ordered structures that
typically are backed by binary trees. Since we are putting pseudorandom
paths into these and looking them up by exact key, we don't need the
ordering, and we're paying a nontrivial cost per insertion.
The existing maps require O(n log n) memory and have O(log n) insertion
and lookup time.
We could instead use unordered maps, which are typically backed by
hashmaps. These require O(n) memory and have O(1) insertion and lookup
time.
On my system this appears to result in a dramatic speedup -- prior to
this patch I was able to delete 400k paths out of 9.5 million over the
course of 34.5 hours. After this patch the same result took 89 minutes.
This result should NOT be taken at face value because the two runs
aren't really comparable; in particular the first started when I had 9.5
million store paths and the seconcd started with 7.8 million, so we are
deleting a different set of paths starting from a much cleaner
filesystem. But I do think it's indicative.
Related: https://github.com/NixOS/nix/issues/9581
This fixes segfaults with nix copy when there was an error processing
addMultipleToStore.
Running with ASAN/TSAN pointed at an use-after-free with threads from
the pool accessing the graph declared in processGraph after the function
was exiting and destructing the variables.
It turns out that if there is an error before pool.process() is called,
for example while we are still enqueuing tasks, then pool.process()
isn't called and threads are still left to run.
By creating the pool last we ensure that it is stopped first before
running other destructors even if an exception happens early.
[ lix porting note: nix does not name threads so the patch has been
adapted to not pass thread name ]
Link: https://git.lix.systems/lix-project/lix/issues/618
Link: https://gerrit.lix.systems/c/lix/+/2355
Commands like `nix flake metadata '.?submodules=1'` ignored the query
part of the URL, while `nix build '.?submodules=1#foo'` did work
correctly because of the presence of the fragment part.
The current instructions for building the Nix manual include a command that doesn't work as described. Specifically:
```
nix build .#nix^doc
```
Running this command results in the error:
```
error: derivation '/nix/store/hddqxzfqgx2fhj8q66ss3idym7pk7aj1-nix-2.26.0pre20250107_383ab87.drv' does not have wanted outputs 'doc'
```
However, this command works if you specify the Nix version explicitly, such as:
```
nix build nix/2.24.11#nix^doc
```
Additionally, these commands are run within the Nix root directory.
However, the nix build .#nix^doc command does work when run from the nixpkgs directory and generates the NixOS manual.
I'm not sure if I'm missing something. Is the `nix^doc` supposed to be added somehow to flake outputs?
The incremental build section does not work since as make has been decommissioned in favor of Meson. Should this be simply deleted?
I'd messed up a rebase in my previous iteration, causing `weakly_canonical` to reappear,
but not trigger a test failure.
These two functions behave similarly when the argument is a path that points to a broken
symlink. `weakly_canonical` would not resolve it because the target doesn't exist, and
`makeParentCanonical` would not resolve it, because it never resolves the final path
element.
This new test case now also tests a valid symlink, "differentiating" the two.
