with position and symbol tables in place we can now shrink Attr by a full
pointer with some simple field reordering. since Attr is a very hot struct this
has substantial impact on memory use, decreasing GC allocations and heap size by
10-15% each. we also get a ~15% performance improvement due to reduced GC
loading.
pure parsing has taken a hit over the branch base because positions are now
slightly more expensive to create, but overall we get a noticeable improvement.
before (on memory-friendliness):
Benchmark 1: nix search --no-eval-cache --offline ../nixpkgs hello
Time (mean ± σ): 6.960 s ± 0.028 s [User: 5.832 s, System: 0.897 s]
Range (min … max): 6.886 s … 7.005 s 20 runs
Benchmark 2: nix eval -f ../nixpkgs/pkgs/development/haskell-modules/hackage-packages.nix
Time (mean ± σ): 328.1 ms ± 1.7 ms [User: 295.8 ms, System: 32.2 ms]
Range (min … max): 324.9 ms … 331.2 ms 20 runs
Benchmark 3: nix eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'
Time (mean ± σ): 2.688 s ± 0.029 s [User: 2.365 s, System: 0.238 s]
Range (min … max): 2.642 s … 2.742 s 20 runs
after:
Benchmark 1: nix search --no-eval-cache --offline ../nixpkgs hello
Time (mean ± σ): 6.902 s ± 0.039 s [User: 5.844 s, System: 0.783 s]
Range (min … max): 6.820 s … 6.956 s 20 runs
Benchmark 2: nix eval -f ../nixpkgs/pkgs/development/haskell-modules/hackage-packages.nix
Time (mean ± σ): 330.7 ms ± 2.2 ms [User: 300.6 ms, System: 30.0 ms]
Range (min … max): 327.5 ms … 334.5 ms 20 runs
Benchmark 3: nix eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'
Time (mean ± σ): 2.330 s ± 0.027 s [User: 2.040 s, System: 0.234 s]
Range (min … max): 2.272 s … 2.383 s 20 runs
this slightly increases the amount of memory used for any given symbol, but this
increase is more than made up for if the symbol is referenced more than once in
the EvalState that holds it. on average every symbol should be referenced at
least twice (once to introduce a binding, once to use it), so we expect no
increase in memory on average.
symbol tables are limited to 2³² entries like position tables, and similar
arguments apply to why overflow is not likely: 2³² symbols would require as many
string instances (at 24 bytes each) and map entries (at 24 bytes or more each,
assuming that the map holds on average at most one item per bucket as the docs
say). a full symbol table would require at least 192GB of memory just for
symbols, which is well out of reach. (an ofborg eval of nixpks today creates
less than a million symbols!)
PosTable deduplicates origin information, so using symbols for paths is no
longer necessary. moving away from path Symbols also reduces the usage of
symbols for things that are not keys in attribute sets, which will become
important in the future when we turn symbols into indices as well.
Pos objects are somewhat wasteful as they duplicate the origin file name and
input type for each object. on files that produce more than one Pos when parsed
this a sizeable waste of memory (one pointer per Pos). the same goes for
ptr<Pos> on 64 bit machines: parsing enough source to require 8 bytes to locate
a position would need at least 8GB of input and 64GB of expression memory. it's
not likely that we'll hit that any time soon, so we can use a uint32_t index to
locate positions instead.
when we introduce position and symbol tables we'll need to do lookups to turn
indices into those tables into actual positions/symbols. having the error
functions as members of EvalState will avoid a lot of churn for adding lookups
into the tables for each caller.
only file and line of the returned position were ever used, it wasn't actually
used a position. as such we may as well use a path+int pair for only those two
values and remove a use of Pos that would not work well with a position table.
a future commit will remove the ability to convert the symbol type used in
bindings to strings. since we only have two users we can inline the error check.
the only use of this function is to determine whether a lambda has a non-set
formal, but this use is arguably better served by Symbol::set and using a
non-Symbol instead of an empty symbol in the parser when no such formal is present.
we don't *need* symbols here. the only advantage they have over strings is
making call-counting slightly faster, but that's a diagnostic feature and thus
needn't be optimized.
this also fixes a move bug that previously didn't show up: PrimOp structs were
accessed after being moved from, which technically invalidates them. previously
the names remained valid because Symbol copies on move, but strings are
invalidated. we now copy the entire primop struct instead of moving since primop
registration happen once and are not performance-sensitive.
Without the change any CA deletion triggers linear scan on large
RealisationsRefs table:
sqlite>.eqp full
sqlite> delete from RealisationsRefs where realisationReference IN ( select id from Realisations where outputPath = 1234567890 );
QUERY PLAN
|--SCAN RealisationsRefs
`--LIST SUBQUERY 1
`--SEARCH Realisations USING COVERING INDEX IndexRealisationsRefsOnOutputPath (outputPath=?)
With the change it gets turned into a lookup:
sqlite> CREATE INDEX IndexRealisationsRefsRealisationReference on RealisationsRefs(realisationReference);
sqlite> delete from RealisationsRefs where realisationReference IN ( select id from Realisations where outputPath = 1234567890 );
QUERY PLAN
|--SEARCH RealisationsRefs USING INDEX IndexRealisationsRefsRealisationReference (realisationReference=?)
`--LIST SUBQUERY 1
`--SEARCH Realisations USING COVERING INDEX IndexRealisationsRefsOnOutputPath (outputPath=?)
If the derivation `foo` depends on `bar`, and they both have the same
output path (because they are CA derivations), then this output path
will depend both on the realisation of `foo` and of `bar`, which
themselves depend on each other.
This confuses SQLite which isn’t able to automatically solve this
diamond dependency scheme.
Help it by adding a trigger to delete all the references between the
relevant realisations.
Fix#5320
The test illustrates failure in issue #5320. Here derivation and
it's built input have identical CA sotre path. As a result we generate
extraneout reference to build input:
$ make installcheck
...
ran test tests/selfref-gc.sh... [PASS]
ran test tests/ca/selfref-gc.sh... [FAIL]
...
deleting '/tmp/.../tests/ca/selfref-gc/store/iqciq1mpg5hc7p6a52fp2bjxbyc9av0v-selfref-gc'
deleting '/tmp/...tests/ca/selfref-gc/store/zh0kwpnirw3qbv6dl1ckr1y0kd5aw6ax-selfref-gc.drv'
error: executing SQLite statement
'delete from ValidPaths where path = '/tmp/.../tests/ca/selfref-gc/store/fsjq0k146r85lsh01l0icl30rnhv7z72-selfref-gc';':
constraint failed (in '/tmp/.../tests/ca/selfref-gc/var/nix/db/db.sqlite')
Otherwise the clang builds fail because the constructor of `SQLiteBusy`
inherits it, `SQLiteError::_throw` tries to call it, which fails.
Strangely, gcc works fine with it. Not sure what the correct behavior is
and who is buggy here, but either way, making it public is at the worst
a reasonable workaround
Don’t say that the derivation is CA as it might happen on a non-ca
derivation too.
Technically we could always recover _something_ for a purely
input-addressed derivation (like we already do when the `ca-derivations`
xp feature isn’t enabled), but it seems better to consistently fail −
the end-result wouldn’t really make sense anyways in most cases.
This ensures that use-sites properly trigger new monomorphisations on
one hand, and on the other hand keeps the main `sqlite.hh` clean and
interface-only. I think that is good practice in general, but in this
situation in particular we do indeed have `sqlite.hh` users that don't
need the `throw_` function.
nix show-config --json was serializing experimental features as ints.
nlohmann::json will automatically use these definitions to serialize
and deserialize ExperimentalFeatures.
Strictly, we don't use the from_json instance yet, it's provided for
completeness and hopefully future use.
Requested by ppepino on the Matrix:
https://matrix.to/#/!KqkRjyTEzAGRiZFBYT:nixos.org/$Tb32BS3rVE2BSULAX4sPm0h6CDewX2hClOTGzTC7gwM?via=nixos.org&via=matrix.org&via=nixos.dev
This adds a new command, :bl, which works like :b but also creates
a GC root symlink to the various derivation outputs.
ckie@cookiemonster ~/git/nix -> ./outputs/out/bin/nix repl
Welcome to Nix 2.6.0. Type :? for help.
nix-repl> :l <nixpkgs>
Added 16118 variables.
nix-repl> :b runCommand "hello" {} "echo hi > $out"
This derivation produced the following outputs:
./repl-result-out -> /nix/store/kidqq2acdpi05c4a9mlbg2baikmzik44-hello
[1 built, 0.0 MiB DL]
ckie@cookiemonster ~/git/nix -> cat ./repl-result-out
hi
Sergei Trofimovich
Yorick
mei (ckie)