More precisely, in concatLists, if all lists except one are empty,
then just return the non-empty list. This reduces the number of list
element allocations by 32% when evaluating a NixOS system
configuration.
This can serve as a generic efficient list builder. For instance, the
function ‘catAttrs’ in Nixpkgs can be rewritten from
attr: l: fold (s: l: if hasAttr attr s then [(getAttr attr s)] ++ l else l) [] l
to
attr: l: builtins.concatLists (map (s: if hasAttr attr s then [(getAttr attr s)] else []) l)
Statistics before:
time elapsed: 1.08683
size of a value: 24
environments allocated: 1384376 (35809568 bytes)
list elements: 6946783 (55574264 bytes)
list concatenations: 37434
values allocated: 1760440 (42250560 bytes)
attribute sets allocated: 392040
right-biased unions: 186334
values copied in right-biased unions: 591137
symbols in symbol table: 18273
number of thunks: 1297673
number of thunks avoided: 1380759
number of attr lookups: 430802
number of primop calls: 628912
number of function calls: 1333544
Statistics after (including new catAttrs):
time elapsed: 0.959854
size of a value: 24
environments allocated: 1010198 (26829296 bytes)
list elements: 1984878 (15879024 bytes)
list concatenations: 30488
values allocated: 1589760 (38154240 bytes)
attribute sets allocated: 392040
right-biased unions: 186334
values copied in right-biased unions: 591137
symbols in symbol table: 18274
number of thunks: 1040925
number of thunks avoided: 1038428
number of attr lookups: 438419
number of primop calls: 474844
number of function calls: 959366
Evaluation of a NixOS configuration spends quite a lot of time in the
"filter" function in Nixpkgs. As implemented in Nixpkgs, this is a
O(n^2) operation, so it's a good candidate for providing a more
efficient (i.e. primop) implementation. Using it gives a ~10% speed
increase and a significant reduction in the number of evaluations.
Statistics before (on a NixOS system config):
time elapsed: 1.3258
size of a value: 24
environments allocated: 1980939 (50127080 bytes)
list elements: 14679308 (117434464 bytes)
list concatenations: 50828
values allocated: 2098938 (50374512 bytes)
attribute sets allocated: 392040
right-biased unions: 186334
values copied in right-biased unions: 591137
symbols in symbol table: 18271
number of thunks: 1645752
number of thunks avoided: 1921196
number of attr lookups: 430798
number of primop calls: 838807
number of function calls: 1930107
Statistics after:
time elapsed: 1.17982
size of a value: 24
environments allocated: 1543334 (39624560 bytes)
list elements: 9612638 (76901104 bytes)
list concatenations: 37434
values allocated: 1854933 (44518392 bytes)
attribute sets allocated: 392040
right-biased unions: 186334
values copied in right-biased unions: 591137
symbols in symbol table: 18272
number of thunks: 1392467
number of thunks avoided: 1507311
number of attr lookups: 430801
number of primop calls: 691600
number of function calls: 1492502
Setting the environment variable NIX_COUNT_CALLS to 1 enables some
basic profiling in the evaluator. It will count calls to functions
and primops as well as evaluations of attributes.
For example, to see where evaluation of a NixOS configuration spends
its time:
$ NIX_SHOW_STATS=1 NIX_COUNT_CALLS=1 ./src/nix-instantiate/nix-instantiate '<nixos>' -A system --readonly-mode
...
calls to 39 primops:
239532 head
233962 tail
191252 hasAttr
...
calls to 1595 functions:
224157 `/nix/var/nix/profiles/per-user/root/channels/nixos/nixpkgs/pkgs/lib/lists.nix:17:19'
221767 `/nix/var/nix/profiles/per-user/root/channels/nixos/nixpkgs/pkgs/lib/lists.nix:17:14'
221767 `/nix/var/nix/profiles/per-user/root/channels/nixos/nixpkgs/pkgs/lib/lists.nix:17:10'
...
evaluations of 7088 attributes:
167377 undefined position
132459 `/nix/var/nix/profiles/per-user/root/channels/nixos/nixpkgs/pkgs/lib/attrsets.nix:119:41'
47322 `/nix/var/nix/profiles/per-user/root/channels/nixos/nixpkgs/pkgs/lib/attrsets.nix:13:21'
...
prevents files from being evaluated and stored as values multiple
times. For instance, evaluation of the ‘system’ attribute in NixOS
causes ‘nixpkgs/pkgs/lib/lists.nix’ to be evaluated 2019 times.
Caching gives a modest speedup and a decent memory footprint
reduction (e.g., from 1.44s to 1.28s, and from 81 MiB to 59 MiB with
GC_INITIAL_HEAP_SIZE=100000 on my system).
brackets, e.g.
import <nixpkgs/pkgs/lib>
are resolved by looking them up relative to the elements listed in
the search path. This allows us to get rid of hacks like
import "${builtins.getEnv "NIXPKGS_ALL"}/pkgs/lib"
The search path can be specified through the ‘-I’ command-line flag
and through the colon-separated ‘NIX_PATH’ environment variable,
e.g.,
$ nix-build -I /etc/nixos ...
If a file is not found in the search path, an error message is
lazily thrown.
write ‘attrs ? a.b’ to test whether ‘attrs’ has an attribute ‘a’
containing an attribute ‘b’. This is more convenient than ‘attrs ?
a && attrs.a ? b’.
Slight change in the semantics: it's no longer an error if the
left-hand side of ‘?’ is not an attribute set. In that case it just
returns false. So, ‘null ? foo’ no longer throws an error.
little RAM. Even if the memory isn't actually used, it can cause
problems with the overcommit heuristics in the kernel. So use a VM
space of 25% of RAM, up to 384 MB.
tree). This saves a lot of memory. The vector should be sorted so
that names can be looked up using binary search, but this is not the
case yet. (Surprisingly, looking up attributes using linear search
doesn't have a big impact on performance.)
Memory consumption for
$ nix-instantiate /etc/nixos/nixos/tests -A bittorrent.test --readonly-mode
on x86_64-linux with GC enabled is now 185 MiB (compared to 946
MiB on the trunk).
improves GC effectiveness a bit more (because a live value doesn't
keep other values in the environment plus the parent environments
alive), and removes the need for copy nodes.
a pointer to a Value, rather than the Value directly. This improves
the effectiveness of garbage collection a lot: if the Value is
stored inside the set directly, then any live pointer to the Value
causes all other attributes in the set to be live as well.
errors with position info.
* For all positions, use the position of the first character of the
first token, rather than the last character of the first token plus
one.
values. This improves sharing and gives another speed up.
Evaluation of the NixOS system attribute is now almost 7 times
faster than the old evaluator.
use site, allowing environments to be stores as vectors of values
rather than maps. This should speed up evaluation and reduce the
number of allocations.
precedence, i.e. `with {x=1;}; with {x=2;}; x' evaluates to 2'.
This has a simpler implementation and seems more natural. There
doesn't seem to be any code in Nixpkgs or NixOS that relies on the
old behaviour.
efficiently. The symbol table ensures that there is only one copy
of each symbol, thus allowing symbols to be compared efficiently
using a pointer equality test.
then the blackhole has to be removed to ensure that repeated
evaluation of the same value gives an assertion failure again rather
than an "infinite recursion" error.
that there are some places in Nixpkgs (php_configurable /
composableDerivation, it seems) that call `derivation' with
incorrect arguments (namely, the `name' attribute missing) but get
away with it because of laziness.
* Removed exprToString and stringToExpr because there is no ATerm
representation to work on anymore (and exposing the internals of the
evaluator like this is not a good idea anyway).
allowed. So `name1@name2', `{attrs1}@{attrs2}' and so on are now no
longer legal. This is no big loss because they were not useful
anyway.
This also changes the output of builtins.toXML for @-patterns
slightly.
broken, but now the evaluator checks for it to prevent Nix
expressions from relying on undefined behaviour. Equality tests are
implemented using a shallow pointer equality test between ATerms.
However, because attribute sets are lazy and contain position
information, this can give false positives. For instance,
previously
let y = {x = 1;}; in y == y
evaluated to true, while the equivalent expression
{x = 1;} == {x = 1;}
evaluated to false. So disallow these tests for now. (Eventually
we may want to implement deep equality tests for attribute sets,
like lib.eqStrict.)
* Idem: disallow comparisons between functions.
* Implemented deep comparisons of lists. This had the same problem as
attribute sets - the elements in the list weren't evaluated. For
instance,
["xy"] == [("x" + "y")]
evaluated to false. Now it works properly.
(which means it can only be defined via "inherit"), otherwise we get
scoping bugs, since __overrides can't be recursive (or at least, it
would be hard).
a rec. This will be very useful to allow end-user customisation of
all-packages.nix, for instance globally overriding GCC or some other
dependency. The // operator doesn't cut it: you could replace the
"gcc" attribute, but all other attributes would continue to
reference the original value due to the substitution semantics of
rec.
The syntax is a bit hacky but this is to allow backwards
compatibility.
in attribute set pattern matches. This allows defining a function
that takes *at least* the listed attributes, while ignoring
additional attributes. For instance,
{stdenv, fetchurl, fuse, ...}:
stdenv.mkDerivation {
...
};
defines a function that requires an attribute set that contains the
specified attributes but ignores others. The main advantage is that
we can then write in all-packages.nix
aefs = import ../bla/aefs pkgs;
instead of
aefs = import ../bla/aefs {
inherit stdenv fetchurl fuse;
};
This saves a lot of typing (not to mention not having to update
all-packages.nix with purely mechanical changes). It saves as much
typing as the "args: with args;" style, but has the advantage that
the function arguments are properly declared (not implicit in what
the body of the "with" uses).
functions that take a single argument (plain lambdas) into one AST
node (Function) that contains a Pattern node describing the
arguments. Current patterns are single lazy arguments (VarPat) and
matching against an attribute set (AttrsPat).
This refactoring allows other kinds of patterns to be added easily,
such as Haskell-style @-patterns, or list pattern matching.
;-)
* Channels: fix channels that are plain lists of derivations (like
strategoxt-unstable) instead of functions (like nixpkgs-unstable).
This fixes the error message "error: the left-hand side of the
function call is neither a function nor a primop (built-in
operation) but a list".
evaluator. This was important because the NixOS expressions started
to hit 2 MB default stack size on Linux.
GCC is really dumb about stack space: it just adds up all the local
variables and temporaries of every scope into one huge stack frame.
This is really bad for deeply recursive functions. For instance,
every `throw Error(format("error message"))' causes a format object
of a few hundred bytes to be allocated on the stack. As a result,
every recursive call to evalExpr2() consumed 4680 bytes. By
splitting evalExpr2() and by moving the exception-throwing code out
of the main functions, evalExpr2() now only consumes 40 bytes.
Similar for evalExpr().
containing functions that operate on the Nix store. One
implementation is LocalStore, which operates on the Nix store
directly. The next step, to enable secure multi-user Nix, is to
create a different implementation RemoteStore that talks to a
privileged daemon process that uses LocalStore to perform the actual
operations.
concatenation and string coercion. This was a big mess (see
e.g. NIX-67). Contexts are now folded into strings, so that they
don't cause evaluation errors when they're not expected. The
semantics of paths has been clarified (see nixexpr-ast.def).
toString() and coerceToString() have been merged.
Semantic change: paths are now copied to the store when they're in a
concatenation (and in most other situations - that's the
formalisation of the meaning of a path). So
"foo " + ./bla
evaluates to "foo /nix/store/hash...-bla", not "foo
/path/to/current-dir/bla". This prevents accidental impurities, and
is more consistent with the treatment of derivation outputs, e.g.,
`"foo " + bla' where `bla' is a derivation. (Here `bla' would be
replaced by the output path of `bla'.)
side should be a path, I guess.
* Handle paths that are in the store but not direct children of the
store directory.
* Ugh, hack to prevent double context wrapping.
derivation attributes to flatten them into strings. This is
possible since string can nowadays be wrapped in contexts that
describe the derivations/sources referenced by the evaluation of the
string.
argument has a valid value, i.e., is in a certain domain. E.g.,
{ foo : [true false]
, bar : ["a" "b" "c"]
}: ...
This previously could be done using assertions, but domain checks
will allow the buildfarm to automatically extract the configuration
space from functions.
"--with-freetype2-library=" + freetype + "/lib"
can now be written as
"--with-freetype2-library=${freetype}/lib"
An arbitrary expression can be enclosed within ${...}, not just
identifiers.
* Escaping in string literals: \n, \r, \t interpreted as in C, any
other character following \ is interpreted as-is.
* Newlines are now allowed in string literals.
expressions that cause an assertion failure (like `assert system ==
"i686-linux"'). This allows all-packages.nix in Nixpkgs to be used
on all platforms, even if some Nix expressions don't work on all
platforms.
Not sure if this is a good idea; it's a bit hacky. In particular,
due to laziness some derivations might appear in `nix-env -qa' but
disappear in `nix-env -qas' or `nix-env -i'.
Commit 5000!
Instead we generate data bindings (build and match functions) for
the constructors specified in `constructors.def'. In particular
this removes the conversions between AFuns and strings, and Nix
expression evaluation now seems 3 to 4 times faster.
The expression `with E1; E2' evaluates to E2 with all bindings in
the attribute set E1 substituted. E.g.,
with {x = 123;}; x
evaluates to 123. That is, the attribute set E1 is in scope in E2.
This is particularly useful when importing files containing lots
definitions. E.g., instead of
let {
inherit (import ./foo.nix) a b c d e f;
body = ... a ... f ...;
}
we can now say
with import ./foo.nix;
... a ... f ...
I.e., we don't have to say what variables should be brought into scope.
print a nice backtrace of the stack, rather than vomiting a gigantic
(and useless) aterm on the screen. Example:
error: while evaluating file `.../pkgs/system/test.nix':
while evaluating attribute `subversion' at `.../pkgs/system/all-packages-generic.nix', line 533:
while evaluating function at `.../pkgs/applications/version-management/subversion/default.nix', line 1:
assertion failed at `.../pkgs/applications/version-management/subversion/default.nix', line 13
Since the Nix expression language is lazy, the trace may be
misleading. The purpose is to provide a hint as to the location of
the problem.
instead of `derivation' triggered a huge slowdown in the Nix
expression evaluator. Total execution time of `nix-env -qa' went up
by a factor of 60 or so.
This scalability problem was caused by expressions such as
(x: y: ... x ...) a b
where `a' is a large term (say, the one in
`all-packages-generic.nix'). Then the first beta-reduction would
produce
(y: ... a ...) b
by substituting `a' for `x'. The second beta-reduction would then
substitute `b' for `y' into the body `... a ...', which is a large
term due to `a', and thus causes a large traversal to be performed
by substitute() in the second reduction. This is however entirely
redundant, since `a' cannot contain free variables (since we never
substitute below a weak head normal form).
The solution is to wrap substituted terms into a `Closed'
constructor, i.e.,
subst(subs, Var(x)) = Closed(e) iff subs[x] = e
have substitution not descent into closed terms,
subst(subs, Closed(x)) = Closed(x)
and otherwise ignore them for evaluation,
eval(Closed(x)) = eval(x).
* Fix a typo that caused incorrect substitutions to be performed in
simple lambdas, e.g., `(x: x: x) a' would reduce to `(x: a)'.
`bla:' is now no longer parsed as a URL.
* Re-enabled support for the `args' attribute in derivations to
specify command line arguments to the builder, e.g.,
...
builder = /usr/bin/python;
args = ["-c" ./builder.py];
...
parser (roughly 80x faster).
The absolutely latest version of Bison (1.875c) is required for
reentrant GLR support, as well as a recent version of Flex (say,
2.5.31). Note that most Unix distributions ship with the
prehistoric Flex 2.5.4, which doesn't support reentrancy.
Nix. This is to prevent Berkeley DB from becoming wedged.
Unfortunately it is not possible to throw C++ exceptions from a
signal handler. In fact, you can't do much of anything except
change variables of type `volatile sig_atomic_t'. So we set an
interrupt flag in the signal handler and check it at various
strategic locations in the code (by calling checkInterrupt()).
Since this is unlikely to cover all cases (e.g., (semi-)infinite
loops), sometimes SIGTERM may now be required to kill Nix.
