Currently, we have a number of assertions in the codebase and certain
assumptions about the AST. These don't always hold, sometimes leading to
a crash in the extractor.
The crashes leave incomplete TRAP files that cannot be imported into the
database.
With this change, we still get those incomplete TRAP files, but we also
get a database in the end (even thoough it is also incomplete as we
cannot import everything).
The `getName` in `Type.qll` was issuing a warning in other generated
classes having a `getName` from a `name` property in `schema.yml`.
To fix the possible inconsistency, `diagnostic_name` is being renamed to
`name` in the schema. Despite the scary doc comment on
`swift::Type::getString` (namely `for use in diagnostics only`), that
seems to be the right generic naming mechanism for types, and it
coincides with the name we were extracting on types with an explicit
`name` property.
In case we find a case where `Type::getString` gives something wrong,
we can probably just patch it on that specific type class.
Visitor code has been split between header and sources to speed up
incremental build. Moreover the code was reorganized using a new `infra`
bazel package (and `visitors` got promoted to a bazel package as well).
This new class encompasses both `AbstractFunctionDecl` and
`AbstractClosureExpr`, together with their common parts (namely
parameters and the body).
`ClosureExpr` and `AutoClosureExpr` got ported to structured C++
generated translation in the process.
This was temporarily broken as we were skipping full emission of all
entities without any valid location.
We now rely on `decl->getDeclContext()->getParentSourceFile()` which is
more robust.
`getLocation()` will now exists for all entities. When there is no
valid location, the location will still not be emitted in the DB, but
on the QL side we will then assign a special `UnknownLocation` with
empty filename and 0 for line/column start/end.
This unknown location is currently emitted (with a unique `@` key) at
the start of every extraction, but we can move it elsewhere (and
possibly in a unique global trap file) at a later stage, possibly after
or when we rework the trap file strategy.
This should solve flakiness that was observed on the control flow tests,
which is probably caused by the `nodes` predicate in the `TestOutput`
class in `ControlFlowGraphImplShared.qll` not able to assign a proper
rank when the node does not have a location.
This adds:
* a base `README.md` file to `codegen`
* module docstrings for the modules in `generators`
* help strings on all command line flags
Moreover some unneeded command line flags (`--namespace`,
`--include-dir` and `--trap-affix`) have been dropped.
