GitLab

Out of curiousity, I entered hadrian's build output directory of a validate flavoured build (so I'm not exactly sure if that's representative) and ran

find . -type f -name "*.o" | xargs du {} | sort -n

These are the top hitters:

1016    ./stage1/compiler/build/GHC/Hs/Decls.o
1024    ./stage0/compiler/build/GHC/Tc/Errors.o
1024    ./stage1/libraries/Cabal/Cabal/build/Distribution/PackageDescription/FieldGrammar.o
1024    ./stage1/utils/haddock/build/Haddock/Backends/LaTeX.o
1072    ./stage0/utils/haddock/build/Haddock/Backends/LaTeX.o
1072    ./stage1/compiler/build/GHC/Core/Lint.o
1072    ./stage1/compiler/build/GHC/StgToCmm/Prim.o
1076    ./stage0/compiler/build/GHC/Driver/Make.o
1076    ./stage1/compiler/build/GHC/CmmToAsm/X86/CodeGen.o
1136    ./stage1/libraries/ghci/build/GHCi/TH/Binary.o
1192    ./stage1/libraries/base/build/Data/Data.o
1236    ./stage1/compiler/build/GHC/Parser/Lexer.o
1272    ./stage1/libraries/Cabal/Cabal/build/Language/Haskell/Extension.o
1308    ./stage0/compiler/build/GHC/Parser/Lexer.o
1324    ./stage0/libraries/Cabal/Cabal/build/Language/Haskell/Extension.o
1348    ./stage0/compiler/build/GHC/Core/Lint.o
1364    ./stage1/compiler/build/GHC/Iface/Ext/Ast.o
1440    ./stage0/compiler/build/GHC/Iface/Ext/Ast.o
1476    ./stage1/ghc/build/GHCi/UI.o
1488    ./stage0/compiler/build/GHC/StgToCmm/Prim.o
1488    ./stage0/compiler/build/GHC/Tc/TyCl.o
1492    ./stage0/compiler/build/GHC/Tc/Module.o
1504    ./stage0/libraries/ghci/build/GHCi/TH/Binary.o
1548    ./stage1/compiler/build/GHC/Hs/Expr.o
1588    ./stage0/compiler/build/GHC/CmmToAsm/X86/CodeGen.o
1608    ./stage1/compiler/build/GHC/Platform/Constants.o
1776    ./stage0/compiler/build/GHC/Hs/Expr.o
1812    ./stage0/compiler/build/GHC/Hs/Decls.o
1824    ./stage0/libraries/Cabal/Cabal/build/Distribution/SPDX/LicenseId.o
1832    ./stage1/libraries/ghci/build/GHCi/Message.o
1880    ./stage0/compiler/build/GHC/Platform/Constants.o
1904    ./stage1/libraries/Cabal/Cabal/build/Distribution/Simple/Setup.o
1940    ./stage0/libraries/ghci/build/GHCi/Message.o
1996    ./stage1/libraries/ghc-prim/build/GHC/Types.o
2036    ./stage0/libraries/Cabal/Cabal/build/Distribution/Simple/Setup.o
2052    ./stage1/libraries/containers/containers/build/Data/Sequence/Internal.o
2064    ./stage1/compiler/build/GHC/Parser.o
2248    ./stage0/compiler/build/GHC/Parser.o
2308    ./stage1/libraries/ghc-prim/build/GHC/Classes.o
2524    ./stage1/libraries/Cabal/Cabal/build/Distribution/SPDX/LicenseId.o
2988    ./stage1/compiler/build/GHC/Builtin/PrimOps.o
3016    ./stage0/compiler/build/GHC/Driver/Session.o
3072    ./stage0/compiler/build/GHC/Builtin/PrimOps.o
3776    ./stage1/libraries/template-haskell/build/Language/Haskell/TH/Syntax.o
3816    ./stage1/compiler/build/GHC/Driver/Session.o
3916    ./stage0/libraries/template-haskell/build/Language/Haskell/TH/Syntax.o
5396    ./stage1/compiler/build/GHC/Hs/Instances.o
5864    ./stage0/compiler/build/GHC/Hs/Instances.o

These are all files taking > 1 MB. That wouldn't be too surprising if most of them made use of deriving, like Instances.o, which is nearly 6MB in size. But some of them don't! Take GHC.Iface.Syntax as an example. Not a single deriving clause, only very few instances. There are quite a few data type defs, but there's not much code. And there aren't many ASSERTs either, so -DDEBUG shouldn't have much of an impact. Note how both the optimised, non-debug stage0 and the -O1 optimised, debug stage1 version of GHC.Iface.Syntax are in that list.

We should investigate this, maybe there are some easy wins.