This also bans various sneaking of negative numbers from the language
into unsuspecting builtins as was exposed while auditing the
consequences of changing the Nix language integer type to a newtype.
It's unlikely that this change comprehensively ensures correctness when
passing integers out of the Nix language and we should probably add a
checked-narrowing function or something similar, but that's out of scope
for the immediate change.
During the development of this I found a few fun facts about the
language:
- You could overflow integers by converting from unsigned JSON values.
- You could overflow unsigned integers by converting negative numbers
into them when going into Nix config, into fetchTree, and into flake
inputs.
The flake inputs and Nix config cannot actually be tested properly
since they both ban thunks, however, we put in checks anyway because
it's possible these could somehow be used to do such shenanigans some
other way.
Note that Lix has banned Nix language integer overflows since the very
first public beta, but threw a SIGILL about them because we run with
-fsanitize=signed-overflow -fsanitize-undefined-trap-on-error in
production builds. Since the Nix language uses signed integers, overflow
was simply undefined behaviour, and since we defined that to trap, it
did.
Trapping on it was a bad UX, but we didn't even entirely notice
that we had done this at all until it was reported as a bug a couple of
months later (which is, to be fair, that flag working as intended), and
it's got enough production time that, aside from code that is IMHO buggy
(and which is, in any case, not in nixpkgs) such as
https://git.lix.systems/lix-project/lix/issues/445, we don't think
anyone doing anything reasonable actually depends on wrapping overflow.
Even for weird use cases such as doing funny bit crimes, it doesn't make
sense IMO to have wrapping behaviour, since two's complement arithmetic
overflow behaviour is so *aggressively* not what you want for *any* kind
of mathematics/algorithms. The Nix language exists for package
management, a domain where bit crimes are already only dubiously in
scope to begin with, and it makes a lot more sense for that domain for
the integers to never lose precision, either by throwing errors if they
would, or by being arbitrary-precision.
Fixes: https://github.com/NixOS/nix/issues/10968
Original-CL: https://gerrit.lix.systems/c/lix/+/1596
Change-Id: I51f253840c4af2ea5422b8a420aa5fafbf8fae75
The code that counts the number of elided attrs incorrectly used the
per-printer "global" attribute counter instead of a counter that
was relevant only to the current attribute set.
This bug flew under the radar because often the attribute sets aren't
nested, not big enough, or we wouldn't pay attention to the numbers.
I've noticed the issue because the difference underflowed.
Although this behavior is tested by the functional test
lang/eval-fail-bad-string-interpolation-4.nix, the underflow slipped
through review. A simpler reproducer would be as follows, but I
haven't added it to the test suite to keep it simple and marginally
faster.
```
$ nix run nix/2.23.1 -- eval --expr '"" + (let v = { a = { a = 1; b = 2; c = 1; d = 1; e = 1; f = 1; g = 1; h = 1; }; b = { a = 1; b = 1; c = 1; }; }; in builtins.deepSeq v v)'
error:
… while evaluating a path segment
at «string»:1:6:
1| "" + (let v = { a = { a = 1; b = 2; c = 1; d = 1; e = 1; f = 1; g = 1; h = 1; }; b = { a = 1; b = 1; c = 1; }; }; in builtins.deepSeq v v)
| ^
error: cannot coerce a set to a string: { a = { a = 1; b = 2; c = 1; d = 1; e = 1; f = 1; g = 1; h = 1; }; b = { a = 1; «4294967289 attributes elided» }; }
```
this needs a string comparison because there seems to be no other way to
get that information out of bison. usually the location info is going to
be correct (pointing at a bad token), but since EOF isn't a token as
such it'll be wrong in that this case.
this hasn't shown up much so far because a single line ending *is* a
token, so any file formatted in the usual manner (ie, ending in a line
ending) would have its EOF position reported correctly.
the parser treats a plain \r as a newline, error reports do not. this
can lead to interesting divergences if anything makes use of this
feature, with error reports pointing to wrong locations in the input (or
even outside the input altogether).
previously we reported the error at the beginning of the binding
block (for plain inherits) or the beginning of the attr list (for
inherit-from), effectively hiding where exactly the error happened.
this also carries over to runtime positions of attributes in sets as
reported by unsafeGetAttrPos. we're not worried about this changing
observable eval behavior because it *is* marked unsafe, and the new
behavior is much more useful.
we already normalize attr order to lexicographic, doing the same for
formals makes sense. doubly so because the order of formals would
otherwise depend on the context of the expression, which is not quite as
useful as one might expect.
the parser modifies its inputs, which means that sharing them between
the error context reporting system and the parser itself can confuse the
reporting system. usually this led to early truncation of error context
reports which, while not dangerous, can be quite confusing.
desugaring inherit-from to syntactic duplication of the source expr also
duplicates side effects of the source expr (such as trace calls) and
expensive computations (such as derivationStrict).
Commit 83c067c0fa changed `builtins.pathExists`
to resolve symlinks before checking for existence. Consequently, if the path
refers to a symlink itself, existence of the target of the symlink (instead of
the symlink itself) was checked. Restore the previous behavior by skipping
symlink resolution in the last component.
for plain inherits this is really just a stylistic choice, but for
inherit-from it actually fixes an exponential size increase problem
during expr printing (as may happen during assertion failure reporting,
on during duplicate attr detection in the parser)
this also has the effect of sorting let bindings lexicographically
rather than by symbol creation order as was previously done, giving a
better canonicalization in the process.
Pretty-print values in the REPL by printing each item in a list or
attrset on a separate line. When possible, single-item lists and
attrsets are printed on one line, as long as they don't contain a nested
list, attrset, or thunk.
Before:
```
{ attrs = { a = { b = { c = { }; }; }; }; list = [ 1 ]; list' = [ 1 2 3 ]; }
```
After:
```
{
attrs = {
a = {
b = {
c = { };
};
};
};
list = [ 1 ];
list' = [
1
2
3
];
}
```
While preparing PRs like #9753, I've had to change error messages in
dozens of code paths. It would be nice if instead of
EvalError("expected 'boolean' but found '%1%'", showType(v))
we could write
TypeError(v, "boolean")
or similar. Then, changing the error message could be a mechanical
refactor with the compiler pointing out places the constructor needs to
be changed, rather than the error-prone process of grepping through the
codebase. Structured errors would also help prevent the "same" error
from having multiple slightly different messages, and could be a first
step towards error codes / an error index.
This PR reworks the exception infrastructure in `libexpr` to
support exception types with different constructor signatures than
`BaseError`. Actually refactoring the exceptions to use structured data
will come in a future PR (this one is big enough already, as it has to
touch every exception in `libexpr`).
The core design is in `eval-error.hh`. Generally, errors like this:
state.error("'%s' is not a string", getAttrPathStr())
.debugThrow<TypeError>()
are transformed like this:
state.error<TypeError>("'%s' is not a string", getAttrPathStr())
.debugThrow()
The type annotation has moved from `ErrorBuilder::debugThrow` to
`EvalState::error`.
Use `diff --color=always` to print colored output for language test
failures. I've also flipped the arguments so that expected lines missing
from the actual output will be marked with a red `-` and additional
lines found in the actual output will be marked with a green `+`.
Previously it was the other way around, which was very confusing.
This extends the `error: cannot coerce a TYPE to a string` message
to print the value that could not be coerced. This helps with debugging
by making it easier to track down where the value is being produced
from, especially in errors with deep or unhelpful stack traces.
The Nix team has requested that this output format remain unchanged.
I've added a warning to the man page explaining that `nix-instantiate
--eval` output will not parse correctly in many situations.
Previously, there were two mostly-identical value printers -- one in
`libexpr/eval.cc` (which didn't force values) and one in
`libcmd/repl.cc` (which did force values and also printed ANSI color
codes).
This PR unifies both of these printers into `print.cc` and provides a
`PrintOptions` struct for controlling the output, which allows for
toggling whether values are forced, whether repeated values are tracked,
and whether ANSI color codes are displayed.
Additionally, `PrintOptions` allows tuning the maximum number of
attributes, list items, and bytes in a string that will be displayed;
this makes it ideal for contexts where printing too much output (e.g.
all of Nixpkgs) is distracting. (As requested by @roberth in
https://github.com/NixOS/nix/pull/9554#issuecomment-1845095735)
Please read the tests for example output.
Future work:
- It would be nice to provide this function as a builtin, perhaps
`builtins.toStringDebug` -- a printing function that never fails would
be useful when debugging Nix code.
- It would be nice to support customizing `PrintOptions` members on the
command line, e.g. `--option to-string-max-attrs 1000`.
