bootstrap: Document that setting `description` changes the version hash

[mikebenfield]

If I change description = in config.toml, the Crate IDs change, which leads to different name mangling. In particular this means it's not possible to use PGO profiles produced by a compiler with one description to build a compiler with a different description. Is this intentional?

I'm not sure if v0 mangling actually has anything to do with this, but that's where it was convenient for me to insert the code to print IDs.

To see this, apply this patch:

From 705dac4190d0e1c67a848b5af4ba6fc51e3aaa56 Mon Sep 17 00:00:00 2001
From: Michael Benfield <mbenfield@google.com>
Date: Wed, 26 Oct 2022 01:11:29 +0000
Subject: [PATCH] print

---
 compiler/rustc_symbol_mangling/src/v0.rs | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/compiler/rustc_symbol_mangling/src/v0.rs b/compiler/rustc_symbol_mangling/src/v0.rs
index ecfe6861e84..ee21131970f 100644
--- a/compiler/rustc_symbol_mangling/src/v0.rs
+++ b/compiler/rustc_symbol_mangling/src/v0.rs
@@ -739,6 +739,9 @@ fn path_crate(self, cnum: CrateNum) -> Result<Self::Path, Self::Error> {
         let stable_crate_id = self.tcx.def_path_hash(cnum.as_def_id()).stable_crate_id();
         self.push_disambiguator(stable_crate_id.to_u64());
         let name = self.tcx.crate_name(cnum);
+        if name.as_str().contains("rustc_serialize") {
+            println!("XXX {} : {:X}", name.as_str(), stable_crate_id.to_u64());
+        }
         self.push_ident(name.as_str());
         Ok(self)
     }
-- 

and make these two files:

% cat configA.toml 
[rust]
description = "A"
new-symbol-mangling = true
~/Code/rust3  [ mbenfield ]
% cat configB.toml 
[rust]
description = "B"
new-symbol-mangling = true

then do

./x.py build --stage 2 --config configA.toml | grep XXX

and

./x.py build --stage 2 --config configB.toml | grep XXX

and you'll get different outputs.

[Mark-Simulacrum]

Yes, this is expected (and largely desired, IMO, though it's tricky to make the right tradeoff here for all use cases). The comment above description intends to make this clear, perhaps could use some clarification - feel free to file a PR there.

I think it could make sense for us to have a way of exposing metadata that's explicitly informational and doesn't affect metadata, but I worry about it ending up misused. The whole point of changing the hash or mangling here is that if you're building with different settings, we really can't guarantee compatibility in any nice way.

[lolbinarycat]

This could be a comment in config.example.toml, maybe?

[PNDTG]

I think it could be mentioned in the docs where it talks more about cargo, or maybe somewhere in the docs that talks about assembly.

[amahuli03]

Hi, I noticed this issue is still unresolved. If it still needs to be fixed, I can work on it. Please assign it to me.

[lolbinarycat]

@anishamahuli in the future, you can use @rustbot claim to assign yourself to an issue.