Annotations not written to interface files

[rl]

Small example:

module C where

data T a = T a

Compile it:

newbie:tests rl$ ~/projects/ndp/ghc/inplace/bin/ghc-stage2 -O2 -c C.hs
newbie:tests rl$ ls -l C.hi
-rw-r--r--  1 rl  rl  485  4 Dec 12:53 C.hi

Add an annotation: {-# ANN type T () #-}. Compile:

newbie:tests rl$ ~/projects/ndp/ghc/inplace/bin/ghc-stage2 -O2 -c C.hs
Loading package ghc-prim ... linking ... done.
Loading package integer-gmp ... linking ... done.
Loading package base ... linking ... done.
Loading package ffi-1.0 ... linking ... done.
newbie:tests rl$ ls -l C.hi
-rw-r--r--  1 rl  rl  485  4 Dec 12:53 C.hi

Note that the interface file hasn't been updated. Remove C.hi and recompile:

newbie:tests rl$ rm C.hi
newbie:tests rl$ ~/projects/ndp/ghc/inplace/bin/ghc-stage2 -O2 -c C.hs
Loading package ghc-prim ... linking ... done.
Loading package integer-gmp ... linking ... done.
Loading package base ... linking ... done.
Loading package ffi-1.0 ... linking ... done.
newbie:tests rl$ ls -l C.hi
-rw-r--r--  1 rl  rl  507  4 Dec 12:54 C.hi

Only now has the annotation been written to C.hi.

[simonpj]

Roman,

Quite right. The relevant code is in MkIface.lhs. Look at the declaration of IfaceDeclExtras:

data IfaceDeclExtras 
  = IfaceIdExtras    Fixity [IfaceRule]
  | IfaceDataExtras  Fixity [IfaceInstABI] [(Fixity,[IfaceRule])]
  | IfaceClassExtras Fixity [IfaceInstABI] [(Fixity,[IfaceRule])]
  | IfaceSynExtras   Fixity
  | IfaceOtherDeclExtras

This collects information that forms part of an Id's fingerprint, even though the information is not serialised along with the Id itself; instead it occurs elsewhere in the interface file. For example, fixity forms part of an Id's fingerprint, but appears in the mi_fixities field of the ModIface.

What we'd forgotten is that the annotations for an Id should form part of its "extras". So the IfaceIdExtras constructor should look like

  IfaceIdExtras Fixity [IfaceRule] [Serialized]

If you make that change, then I think the knock-on changes (notably filling in that field) will be obvious.

Could you have a go at doing this? Thanks

Simon

[errge]

(In #7867) Nevermind my previous comment, I found the relevant bug:  http://hackage.haskell.org/trac/ghc/ticket/3725

OK, I will continue by investigating if it's possible to get this stuff from template haskell and if not, how hard would be to add it...

I'll get back to this ticket if I have or need further info.