Add a --disable-large-memory-space configure option

GHC recently started using large (1TB) address space on 64 bit systems.

This feature is enabled by default but does not work on arm64-linux or amd64-solaris.

Even just for testing, it would be nice to be able to disable this.

Trac metadata

Trac field	Value
Version	7.11
Type	FeatureRequest
TypeOfFailure	OtherFailure
Priority	normal
Resolution	Unresolved
Component	Build System
Test case
Differential revisions
BlockedBy
Related
Blocking
CC
Operating system
Architecture

changed weight to 5

added Tfeature request Trac import labels

The truth table for enabling USE_LARGE_ADDRESS_SPACE should be:

   sizeof(void*)             command line            USE_LARGE_ADDRESS_SPACE
   -------------------------------------------------------------------------
        4                         -                        undefined
        4           --enable-large-address-space           undefined
        4           --disable-large-address-space          undefined
        8                         -                           1
        8           --enable-large-address-space              1
        8           --disable-large-address-space          undefined

Trac metadata

Trac field	Value
Differential revisions	- → D1170

mentioned in commit fba724cc

closed

Trac metadata

Trac field	Value
Resolution	Unresolved → ResolvedFixed

changed milestone to %8.0.1

Milestone renamed

This code is very hard to read, and the behavior very hard to intuit.

• Folding together the =yes and no-flag case *outside* the "$ac_cv_sizeof_void_p" -eq 8 branch falsely implies the default is unconditionally/constantly true.

• =yes will be silently ignored on non-Darwin if the test doesn't pass.

• Ignoring --enable/disable-large-address-space on 32-bit is more harmful than helpful. The fact is, most (all?) people blindly cargo cult and copy paste configure flags, and this will give them a false intuition. (Postel's law hurts us by hiding bugs with gratuitous complexity.) Crucially and unlike many such similar things, it's not necessary for backwards compatibility since this flag is new.

Granted, I could be persuaded on the 3rd point that on 32-bit the large and small address spaces both morally exist but are the same, rather than the large one not existing, so the laxer handling of the flags is justified. But, at the very least a warning is due on the ignored/superfluous flags, and the code restructured to properly separate sanitizing user input, checking whether expectations are met, and providing a default value.

Sorry to lash out on random 3 year old issue, but this is a perfect example of the sort of "diff golf" / incremental change that leads to impenetrable and every-growing configure scripts (or build systems in general). 5 minutes of what might seem silly OCD by the original author cleaning up what's there can save far more time for the next person. And in fact, writing a truth table a is great first step, without which I would have spent longer being confused.

The new Hadrian build system has been merged. Relabeling the tickets concerning the legacy make build system to prevent confusion.

Trac metadata

Trac field	Value
Component	Build System → Build System (make)

added Pnormal label