hs-boot modification doesn't induce recompilation

Consider these modules:

X.hs-boot

module X where

Y.hs

module Y where
import {-# SOURCE #-} X
data T = T
data S = S T

X.hs

module X where
import Y

ghc --make Y.hs will compile fine. Now, modify X.hs-boot to add a data T (which will cause an ambiguous identifier error in Y.hx. Run ghc --make Y.hs and nothing happens!

(You might also notice something else a bit funny, which is that X.hs gets compiled, even though it's not directly in the import chain of Y. I think this might be intentional but I'm not sure.)

Discovered this while poking around #10182 (closed).

mentioned in issue #10182 (closed)

changed weight to 5

added Tbug Trac import labels

changed the description

I've diagnosed why this is occurring.

GhcMake does some fancy business to determine if an object is stable. Essentially, object stability is a timestamped based check which tries to let GhcMake avoid recompiling source files which didn't change at all.

However, stability is computed on a **per module name** basis, preferentially checking to see if hs files have been updated. This means that if you modify an hs-boot file but not the hs file, GHC will still consider it "stable" because the stability check was done on hs; and will proceed to tell the one-shot compiler that the file was not modified.

I'm not sure what the correct solution to this problem is.

You might also notice something else a bit funny, which is that X.hs gets compiled, even though it's not directly in the import chain of Y. I think this might be intentional but I'm not sure.

This is because there is explicit logic in GhcMake to make this be the case.

mentioned in commit 06d46b1e

mentioned in issue #11013

Related: #11013

assigned to @ezyang

mentioned in issue #11556 (closed)

added hs-boot label

added Pnormal label

mentioned in issue #24779 (closed)

I can't reproduce this as of 9.10.1.