Elaborate on dataflow outputs for region constraints

src/compiler-debugging.md

@@ -244,6 +244,18 @@ $ dot -T pdf maybe_init_suffix.dot > maybe_init_suffix.pdf
 $ firefox maybe_init_suffix.pdf # Or your favorite pdf viewer
 ```
 
+Graphviz also comes with a preprocessor program, 
+[`unflatten`](https://graphviz.org/docs/cli/unflatten/), that
+sometimes helps making the outputs look less oddly spread out. It reads
+a dot file and outputs another dot file, so you can use it in a pipe,
+e.g:
+```
+$ unflatten mir_dump/*.foo.-------.nll.0.regioncx.all.dot | dot -Tpdf  -o foo-outlives.pdf
+```
+
+This is particularly useful for complicated region outlives graphs from 
+[MIR Dataflow](mir/dataflow.md#graphviz-diagrams).
+
 ## Narrowing (Bisecting) Regressions
 
 The [cargo-bisect-rustc][bisect] tool can be used as a quick and easy way to

src/mir/dataflow.md

@@ -217,6 +217,26 @@ the example below:
 
 ![A graphviz diagram for a dataflow analysis](../img/dataflow-graphviz-example.png)
 
+### Region Constraint Graphs and Their Strongly Connected Components
+
+![A graph showing a small number of regions with their outlives relations](../img/region-graphviz.png)
+
+With `-Z dump-mir-graphviz=yes`, you will also get Graphviz files for the outlives constraints
+of the MIR bodies you asked for, as well as the strongly connected components (SCCs) on them.
+They are available as
+`mir_dump/rs-file-name.function-name.-------.nll.0.regioncx.all.dot` and
+`mir_dump/rs-file-name.function-name.-------.nll.0.regioncx.scc.dot` respectively. For both
+graphs, named region variables will be shown with their external name (such as `'static`)
+shown in parenthesis. For region inference variables in universes other than the root universe,
+they will be shown as `/U13` (for universe 13). In the region graph, edges are labelled with
+the MIR locations where the relationship holds, or `All` if it's everywhere.
+
+![A graph showing a small number of strongly connected components on the region-outlives-graph above](../img/scc-graphviz.png)
+
+**Note:** There are implicit edges from `'static` to every region, but those are not rendered
+in the region graph to avoid clutter. They _do_ however show up in the SCC graph. This is why there are outgoing edges from SCC(5) in the SCC graph above
+that do not seem to have corresponding edges in the region outlives graph above.
+
 ["gen-kill" problems]: https://en.wikipedia.org/wiki/Data-flow_analysis#Bit_vector_problems
 [*Static Program Analysis*]: https://cs.au.dk/~amoeller/spa/
 [Debugging MIR]: ./debugging.html