rust/tests/ui/mir-dataflow/README.md

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

54 lines
2.5 KiB
Markdown
Raw Normal View History

This directory contains unit tests for the MIR-based dataflow
analysis.
These unit tests check the dataflow analysis by embedding calls to a
special `rustc_peek` intrinsic within the code, in tandem with an
attribute `#[rustc_mir(rustc_peek_maybe_init)]` (\*). With that
attribute in place, `rustc_peek` calls are a signal to the compiler to
lookup the computed dataflow state for the Lvalue corresponding to the
argument expression being fed to `rustc_peek`. If the dataflow state
for that Lvalue is a 1-bit at that point in the control flow, then no
error is emitted by the compiler at that point; if it is a 0-bit, then
that invocation of `rustc_peek` will emit an error with the message
"rustc_peek: bit not set".
2016-05-20 11:12:02 +00:00
(\*): Or `#[rustc_mir(rustc_peek_maybe_uninit)]`, and perhaps other
variants in the future.
The end effect is that one can write unit tests for MIR dataflow that
perform simple-queries of the computed dataflow state, and the tests
should be able to be robust in the face of changes to how MIR is
represented or constructed.
----
Sometimes understanding the dataflow results is difficult without
looking at the actual MIR control-flow graph being processed with the
corresponding GEN and KILL sets.
For a graphviz-rendering with dataflow annotations, add the attribute
`#[rustc_mir(borrowck_graphviz_postflow="/path/to/suffix.dot")]` to
the function in question. (You can change the content of
`"suffix.dot"` to control the filenames used for the output). This
will generate a separate file for each dataflow analysis, adding a
prefix (corresponding to the name of the analysis) to the filename in
each generated output path.
* For example, the above attribute will currently cause two files to
be generated: `/path/to/maybe_init_suffix.dot` and
`/path/to/maybe_uninit_suffix.dot`.
* The generated `.dot` file shows both the computed dataflow results
on *entry* to each block, as well as the gen- and kill-sets that
were so-called "transfer functions" summarizing the effect of each
basic block.
* (In addition to the `borrowck_graphviz_postflow` attribute-key
noted above, there is also `borrowck_graphviz_preflow`; it has the
same interface and generates the same set of files, but it renders
the dataflow state after building the gen- and kill-sets but
*before* running the dataflow analysis itself, so each entry-set is
just the initial default state for that dataflow analysis. This is
less useful for understanding the error message output in these
tests.)