Use more fine grained locks for the dep graph

[Zoxc]

Split out from #61845.

r? @michaelwoerister cc @aturon

[Zoxc]

@michaelwoerister I'm guessing you're the person to ask about this.

[michaelwoerister]

At some point we had debug output for incremental compilation that was using this. If it is really unused now, you can remove it.

[rust-highfive]

The job x86_64-gnu-llvm-6.0 of your PR failed (raw log). Through arcane magic we have determined that the following fragments from the build log may contain information about the problem.

Click to expand the log.

2019-08-20T19:20:07.0651330Z ##[command]git remote add origin https://github.com/rust-lang/rust
2019-08-20T19:20:07.0864563Z ##[command]git config gc.auto 0
2019-08-20T19:20:07.1299588Z ##[command]git config --get-all http.https://github.com/rust-lang/rust.extraheader
2019-08-20T19:20:07.1362515Z ##[command]git config --get-all http.proxy
2019-08-20T19:20:07.1516728Z ##[command]git -c http.extraheader="AUTHORIZATION: basic ***" fetch --force --tags --prune --progress --no-recurse-submodules --depth=2 origin +refs/heads/*:refs/remotes/origin/* +refs/pull/63756/merge:refs/remotes/pull/63756/merge
---
2019-08-20T19:20:41.2725184Z do so (now or later) by using -b with the checkout command again. Example:
2019-08-20T19:20:41.2725230Z 
2019-08-20T19:20:41.2725436Z   git checkout -b <new-branch-name>
2019-08-20T19:20:41.2725475Z 
2019-08-20T19:20:41.2725522Z HEAD is now at 5ca57c7a4 Merge 80d345747ebb3d890aa08021b969cd24fd7cc0fa into 5a56e05abd34e1936df74625c1f40cb6fee0cd4a
2019-08-20T19:20:41.2892821Z ##[section]Starting: Collect CPU-usage statistics in the background
2019-08-20T19:20:41.2895528Z ==============================================================================
2019-08-20T19:20:41.2895586Z Task         : Bash
2019-08-20T19:20:41.2895632Z Description  : Run a Bash script on macOS, Linux, or Windows
---
2019-08-20T19:30:46.9097528Z    Compiling arena v0.0.0 (/checkout/src/libarena)
2019-08-20T19:30:50.0986540Z    Compiling syntax_pos v0.0.0 (/checkout/src/libsyntax_pos)
2019-08-20T19:30:57.1455584Z    Compiling rustc_errors v0.0.0 (/checkout/src/librustc_errors)
2019-08-20T19:31:12.5147916Z    Compiling fmt_macros v0.0.0 (/checkout/src/libfmt_macros)
2019-08-20T19:31:35.6448390Z error[E0599]: no method named `borrow` found for type `dep_graph::graph::CurrentDepGraph` in the current scope
2019-08-20T19:31:35.6448944Z    --> src/librustc/dep_graph/graph.rs:612:41
2019-08-20T19:31:35.6449175Z     |
2019-08-20T19:31:35.6449528Z 612 |             debug_assert!(!data.current.borrow().node_to_node_index.contains_key(dep_node));
2019-08-20T19:31:35.6450023Z ...
2019-08-20T19:31:35.6450023Z ...
2019-08-20T19:31:35.6450324Z 942 | pub(super) struct CurrentDepGraph {
2019-08-20T19:31:35.6450655Z     | --------------------------------- method `borrow` not found for this
2019-08-20T19:31:35.6461201Z     |
2019-08-20T19:31:35.6461539Z     = help: items from traits can only be used if the trait is in scope
2019-08-20T19:31:35.6461860Z     = note: the following trait is implemented but not in scope, perhaps add a `use` for it:
2019-08-20T19:31:35.6462102Z             `use std::borrow::Borrow;`
2019-08-20T19:31:50.7942952Z error: aborting due to previous error
2019-08-20T19:31:50.7946791Z 
2019-08-20T19:31:50.7956209Z For more information about this error, try `rustc --explain E0599`.
2019-08-20T19:31:50.9721881Z error: Could not compile `rustc`.
---
2019-08-20T19:32:39.4985343Z == clock drift check ==
2019-08-20T19:32:39.5007344Z   local time: Tue Aug 20 19:32:39 UTC 2019
2019-08-20T19:32:39.6574654Z   network time: Tue, 20 Aug 2019 19:32:39 GMT
2019-08-20T19:32:39.6575475Z == end clock drift check ==
2019-08-20T19:32:40.7149248Z ##[error]Bash exited with code '1'.
2019-08-20T19:32:40.7189082Z ##[section]Starting: Checkout
2019-08-20T19:32:40.7191497Z ==============================================================================
2019-08-20T19:32:40.7191574Z Task         : Get sources
2019-08-20T19:32:40.7191645Z Description  : Get sources from a repository. Supports Git, TfsVC, and SVN repositories.

I'm a bot! I can only do what humans tell me to, so if this was not helpful or you have suggestions for improvements, please ping or otherwise contact @TimNN. (Feature Requests)

[michaelwoerister]

Thanks for the PR, @Zoxc!

The changes look correct to me. Do we have evidence that they improve performance?

[Zoxc]

The changes look correct to me. Do we have evidence that they improve performance?

This PR probably doesn't help. It's just a prerequisite for other stuff which does improve performance.

[JohnCSimon]

Ping from triage:
@Zoxc @michaelwoerister - this has sat idle for over nearly two weeks. What else is there to be done for this PR?
Thanks!

[michaelwoerister]

I've taken a look at the code and it looks correct to me, so now it's more of a design question whether we want to do this or not. Since you seem to be doing an audit on all the locking going, @aturon, I'll let you give the go-ahead.

r? @aturon

[JohnCSimon]

Ping from triage
@aturon Can you please review this PR? Thanks.

Thanks.

[JohnTitor]

Ping from triage: @aturon could you review this PR?

[joelpalmer]

Ping from Triage: Pinging @aturon one more time.
CC: @Dylan-DPC

[nikomatsakis]

r? @nikomatsakis

@aturon is stepping back from this role, we should reassign all PRs assigned to them.

[nikomatsakis]

Looks good overall. I have one nit, which is a request to document the overall pattern on CurrentDepGraph.

In general, we discussed this PR a bit in our "parallel sync meeting" today and the general conclusion was that -- in the future -- it might be nice to extract the CurrentDepGraph code into its own module, so that it can have a smaller "abstraction boundary", but since this is part of a larger series of PRs it doesn't make sense to do it now.

We also mentioned using SeqCst everywhere unless there's a strong reason to do otherwise (which doesn't seem to apply here, but @Mark-Simulacrum already noted that here).

[nikomatsakis]

The comment below is great -- but it'd be good to document just a bit more (this is pre-existing, as the code wasn't all that thoroughlly commented before =)

I'd say something like:

---

Encodes the rustc dependency graph. The nodes are identified by an index (DepNodeIndex); the data for each node is stored in its DepNodeData, found in the data field.

We never remove nodes from the graph: they are only added. Most of the time, the node is mapped 1-to-1 with some DepNode, which basically identifies a query. When such nodes are allocated, we add the DepNod into the node_to_node_index map and allocate a fresh node index.

This struct uses two locks internally: the data and node_to_node_index field are locked separately. Operations that begin with a DepNodeIndex typically just access the data field.

The only operation that must manipulate both fields is adding new nodes, in which case we first acquire the node_to_node_index lock and then, once a new node is to be inserted, acquire the lock on data.

(We also have a field fields that use atomics: these are simple counters that are used for profiling and debugging and are not used with debug_assertions.)

[Zoxc]

I added that, but removed the incorrect parts =P

[nikomatsakis]

r=me with comment + SeqCst nits applied =)

[bors]

☔ The latest upstream changes (presumably #65209) made this pull request unmergeable. Please resolve the merge conflicts.

[Zoxc]

@bors r=nikomatsakis

[bors]

📌 Commit b6a5740 has been approved by nikomatsakis

[bors]

⌛ Testing commit b6a5740 with merge c8fa82c...

[bors]

☀️ Test successful - checks-azure
Approved by: nikomatsakis
Pushing c8fa82c to master...