Allow interleaving of STDOUT and STDERR check lines. Put STDOUT lines
after the line they're attached to, and STDERR lines before. If no
STDOUT check line is attached to any line, then put them all at the end
of the file instead.
This is intended to better handle the case where stdout contains
unreplaced mentions of line numbers, and also reflects that stdout is
typically a consequence of the test rather than commentary on it, so
placing it after the test seems likely to read better.
The intent of this change is that instead of paths looking like `explorer/testdata/foo/bar.carbon` (repo-relative), they're now just `bar.carbon` (local). The consequence is that paths should be a bit more durable in various environments, and just paths should be shorter and easier to read.
The explorer's prelude is an exception to this since it comes from data, rather than being the test target. Due to the change in approaches, it needs the regex again.
Uses #2829
Isolating test execution time (no build time included):
- Linux, `lit` test-per-file: Elapsed time: 28.214s, Critical Path: 13.97s
- Linux, `cc_test`-per-file: Elapsed time: 11.534s, Critical Path: 6.05s
- Linux, merged `cc_test` with 50 shards: Elapsed time: 11.677s, Critical Path: 11.17s
- Mac, `lit` test-per-file: Elapsed time: 295.686s, Critical Path: 20.00s
- Mac, `cc_test`-per-file: Elapsed time: 55.788s, Critical Path: 3.81s
- Mac, merged `cc_test` with 50 shards: Elapsed time: 16.269s, Critical Path: 7.54s
In GH actions:
- [Before](https://github.com/carbon-language/carbon-lang/actions/runs/4866602695/jobs/8678306144?pr=2799):
- test / test (ubuntu-22.04, fastbuild) (pull_request_target) Successful in 20m
- test / test (ubuntu-22.04, opt) (pull_request_target) Successful in 15m
- test / test (macos-12, fastbuild) (pull_request_target) Successful in 36m
- test / test (macos-12, opt) (pull_request_target) Successful in 21m
- [After](https://github.com/carbon-language/carbon-lang/actions/runs/4875154751/jobs/8697004066?pr=2811):
- test / test (ubuntu-22.04, fastbuild) (pull_request_target) Successful in 10m
- test / test (ubuntu-22.04, opt) (pull_request_target) Successful in 9m
- test / test (macos-12, fastbuild) (pull_request_target) Successful in 12m
- test / test (macos-12, opt) (pull_request_target) Successful in 9m
I'm still leaving a handful of `lit` tests to test end-to-end binary execution. This is why testdata directories are split (`lit` tests next to the `explorer` binary, the `cc_test`s next to `ParseAndExecute`).
This PR is making two main changes to the Explorer and Toolchain:
- Replace the `is` keyword in `where SomeType is SomeInterface` with `impls`, so it is `where SomeType impls SomeInterface`
- Rewrite uses of the "impls" to something else to avoid, frequently "`impl` declarations" or "implementations", to avoid confusion with the `impls` keyword.
---------
Co-authored-by: Geoff Romer <gromer@google.com>
The strategy that we use for now to support template instantiation is to check the impl declaration as if it were a generic, but to defer all checking of the impl definition until we see a use in which all template parameters have arguments. At that point, we clone the impl definition and type-check the whole thing, with constant values set on the template parameters corresponding to the given arguments.
No caching of template instantiations is performed yet; each time we form a reference to a template instantiation, we instantiate it afresh. We also don't implement the name lookup rule from #949 yet; lookups during template instantiation look only in the actual type and not in the constraint.
Depends on #2699
As described in [the generics design](https://github.com/carbon-language/carbon-lang/blob/trunk/docs/design/generics/details.md#type-structure-of-an-impl-declaration), `impl` declarations are prioritized by type structure. Given two `impl` declarations that match a `type as interface` query, the one that describes the longest prefix of the query without using placeholders is preferred.
We implement this by putting all impls in a total order, first by type structure equivalence classes and then by lexical order. When matching an impl, we walk this total order, and stop once we find a match and reach the end of its equivalence class.
Equivalence classes are determined by finding the locations of the "holes" (the positions where deduced parameters appear) within the type structure, viewed as a tree. Two impls are in the same equivalence class if their holes are in the same place, and equivalence classes are ordered based on a reverse lexicographical ordering of their holes.
Explorer doesn't keep the `Bindings` list for a parameterized type in any particular order, but the type structure rule requires that we consider them in lexical order. In order to support this, we now track an index on the declared parameters of each generic. This is a simple numbering of enclosing generic parameters, both on that generic and on all lexically enclosing generics.
Co-authored-by: Jon Ross-Perkins <jperkins@google.com>
Detect when evaluating an impl recursively tries to evaluate the same impl for the same or a more complex set of parameters.
In order to perform the check after we have tested that the type structure matches and before we check that constraints are recursively satisfied, argument deduction is extended to check structural matching properties earlier.
This requires us to separate match failures into two kinds: hard failures that produce errors that should never be swallowed, and soft failures such as a missing impl that lead us to merely discard an impl as a candidate. A flag has been added to `ImplScope` and `ArgumentDeduction` to specify whether soft failures should produce an error message or not.
Also make minor updates to the skeletal design in
docs/design/name_lookup.md following #2113, as there are no longer any prelude names that are made available to unqualified name lookup by default.
Add `type` to the keyword list in
docs/design/lexical_conventions/words.md, following #2360.
Makes explorer/testdata/assoc_const/rewrite_large_type.carbon NOAUTOUPDATE and no-trace because otherwise it takes ~130s to run. With this it's sub-second, explorer is just dumping a lot of trace output (maybe still something to fix).
When resolving a constraint type, compute the fixed point of the specified rewrite constraints, as requested by @josh11b in review of #2173 and tentatively agreed as our direction. If no such fixed point exists, detect that situation and diagnose the problem.
The algorithm used here is to iteratively apply all rewrites to each rewrite constraint until either one of them refers to itself, which indicates there's a cycle in the rewrite graph, or the set of rewrites converges.
This algorithm has some pathological inputs in which the runtime can grow exponentially in the size of the input (indeed, the fully-rewritten set of rewrites can grow exponentially in the size of the input, so this is unavoidable), so an iteration limit is also provided.
Within the type of an associated constant, references to `.Self` should resolve symbolically to that associated constant as a `GenericBinding` so that it can be substituted for the actual value when using its type. However, when the associated constant is referenced from elsewhere in the same interface, the value we want is a symbolic value naming the constant as a member of `Self`.
Prior to this change, declarations like `let N:! X(.Self) where .(X(.Self).Y) == 5;` have the surprising behavior of the two `.Self` expressions resolving to two different symbolic values. Fix this by forcing the inner one to have the same symbolic value as the outer one, albeit with a different type.
Instead of checking and resolving rewrites eagerly, defer doing so until a constraint is applied to a binding. Actual resolution of rewrites is not yet implemented, so this is mostly just refactoring, but the eventual goal is to support things like `(Constraint where .T = .U) & (Constraint where .U = V)` (however they are reached) by applying one rewrite to the other, as agreed with @josh11b in discussion of #2173.
One nice consequence is that substitution into a constraint type no longer has a failure path, so substitution is once again not able to fail.
Per recent discussion, in `... where .A = B`, require that `B` is implicitly convertible to the type of `A` immediately, rather than treating that as part of the criteria that a type must satisfy to satisfy the resulting constraint. Extend the implementation of implicit conversion so that conversion of a type to a constraint checks that the type satisfies the constraint.
As part of implementing this, stop duplicating rewrite constraints as equality constraints. Instead, when checking that a constraint is satisfied, check both its equality constraints and its rewrite constraints. This fixes an infinite recursion that would otherwise be caused by this change, and is also a necessary prerequisite for applying rewrite constraints to equality constraints, where we would otherwise collapse the implied equality constraints to a tautological `V == V` constraint.
In passing, make ErrorBuilder support building the error message more incrementally and use that to improve diagnostics for mismatched values with equality constraints.
Add checking that a type only satisfies a constraint if it satisfies all of that constraint's equality and rewrite constraints.
Enforce the rule that `=` must be used in impls when specifying associated constant values rather than `==`.
Turn off the pre-#2173 single-step equality behavior. This is getting somewhat ahead of the approved design, but it's a one-line change to restore the old behavior.
Fix `CARBON_CHECK` to handle top-level `,`s in its argument, such as may happen in template argument lists, as this change introduces such a check.
There are lots of ways a declaration can be used before we have the information necessary to handle that use. Issue diagnostics for these.
Interleave declaration and type-checking of global declarations so that declaring a later declaration can depend on the results of type-checking an earlier one.
Incorporates tests added in #2266.
Fixes#1394, fixes#1395, fixes#1396.
Co-authored-by: pmqtt <51272730+pmqtt@users.noreply.github.com>
Co-authored-by: Jon Ross-Perkins <jperkins@google.com>
Adds a simple script to merge stdout/stderr and put on labels. This is hidden to the RUN line using lit.cfg.py.
This is my solution to addressing how errors printed by the toolchain break sorting of stdout output; just put stdout first. We could also have toggles for interleaving output or such, which might help test whether we do it properly.
This also moves some previous-distributed replacement logic into lit_autoupdate_base.py: I think having that adjacent to lit.cfg.py is probably the better choice, and it reduces duplication in toolchain scripts. It happens here because I need to change the resulting commands to include the merge.
This change left a test broken and removed a fair amount of test coverage for unrelated features, and after further discussion it's not clear that this is the direction we want to go in.
This reverts commit 6cf2272bbe.
I've refactored the script in order to make it work in more contexts, which is why the delta is lost. I've actually refactored a significant amount with the intent of making the logic easier to understand, because I was also adjusting bits of it.
Some key notes:
- Removes the multi-pass update that was dealing with unfixed line numbers in explorer (I think the current script should work in one pass)
- Fixed explorer to handle multiple line numbers on the same line (turns out we can rely on local format for line numbers).
- Using execv instead of imports because making Python imports work in a setup like this feels like it's not worth it; only a nuisance.
- Adding __init__.py to satisfy mypy, which otherwise considers the lit_autoupdate.py scripts to be issues.
- Using py because I was thinking sh would be more platform-dependent. py should port better to Windows.
- Getting rid of [[ID#]] capture groups in the semantics-ir tests because with the autoupdate it's kind of moot (also, hard to autogenerate the pairs without relying on the %### value).
Note this does mean tests switch to more of a "make a change, see which tests change" setup. I don't know that that's a _bad_ thing though -- it's pretty much how tests are being written right now, which is why I went down this rabbit hole. It's a nuisance to make a change then _manually_ have to update a bunch of code.
My intent is to use this for to convert parse-tree tests to lit, but I wanted to do this with _existing_ tests first as a proof of concept and to make sure there's agreement.
Implements basic support for rewrite constraints as proposed in #2173.
Support for associated constants and complex constraints in general is also made more robust: argument deduction now properly computes and substitutes witnesses, and interface declarations track a more correct description of the constraint that they introduce than the one we previously built.
Change explorer's handling of witnesses to track them wherever possible. Associated constant support needs witnesses to be available much more pervasively.
The main changes here are:
- `Substitute` now takes a set of bindings covering both `Value`s for `GenericBinding`s and `Witness`es for `ImplBinding`s, and substitutes both.
- Witnesses are now tracked within `ConstraintType`s. The `.Self` type has an `ImplBinding` that self-references from parts of the constraint to other parts of the constraint can use to access the witness for the constraint.
- The constraint type for an `impl` and for a `GenericBinding` are now stored in pre-substituted form, with references to the actual constrained type rather than symbolic references to `.Self`. This results in minor changes in diagnostic text.
Argument deduction still performs substitutions without remapping witnesses.
On #2224 @zygoloid pointed out we needed --implicit-check-not to ensure we were correctly matching output. This is the standard way we're writing explorer tests, so I was looking at unifying our lit approaches.
This is one take on it, making more use of substitutions to bring various testing into alignment, as well as symlinks to avoid config skew (maybe I'll eventually figure out a better solution than symlinks).
Makes a couple small fixes in explorer to remove end-of-line whitespace on output.
Basic support for declaring, specifying the values of, and using associated constants.
This is incomplete in various ways. For example, when checking whether a type satisfies a constraint, there is no check that its associated constants match those in the constraint, and name lookup into a value whose type is an associated constant is not supported yet.
Co-authored-by: Jon Ross-Perkins <jperkins@google.com>
Richard Smith
josh11b
Jon Ross-Perkins
