GitLab

I'm not sure if this could lead to a crash or not, but it upsets +RTS -DS
Might be related to #4265

Stops user-installed packages breaking the build

I got fed up with the constant bogus output from the debugging memory
allocator in RtsUtils.c.  One problem is that we allocate memory in
constructors that then isn't tracked, because the debugging allocator
hasn't been initialised yet.

The bigger problem is that for a given piece of leaking memory it's
impossible to find out where it was allocated; however Valgrind gives
output like this:

==6967== 8 bytes in 1 blocks are still reachable in loss record 1 of 7
==6967==    at 0x4C284A8: malloc (vg_replace_malloc.c:236)
==6967==    by 0x4C28522: realloc (vg_replace_malloc.c:525)
==6967==    by 0x6745E9: stgReallocBytes (RtsUtils.c:213)
==6967==    by 0x68D812: setHeapAlloced (MBlock.c:91)
==6967==    by 0x68D8E2: markHeapAlloced (MBlock.c:116)
==6967==    by 0x68DB56: getMBlocks (MBlock.c:240)
==6967==    by 0x684F55: alloc_mega_group (BlockAlloc.c:305)
==6967==    by 0x6850C8: allocGroup (BlockAlloc.c:358)
==6967==    by 0x69484F: allocNursery (Storage.c:390)
==6967==    by 0x694ABD: allocNurseries (Storage.c:436)
==6967==    by 0x6944F2: initStorage (Storage.c:217)
==6967==    by 0x673E3C: hs_init (RtsStartup.c:160)

which tells us exactly what the leaking bit of memory is.  So I don't
think we need our own debugging allocator.

Hack: dblatex normalises the name of the main input file using
os.path.realpath, which means that if we're in a linked build tree,
it find the real source files rather than the symlinks in our link
tree. This is fine for the static sources, but it means it can't
find the generated sources.

We therefore also generate the main input file, so that it really
is in the link tree, and thus dblatex can find everything.

It's not just when cleaning!

Because rtsIsProfiled is only available in stage2

This was leading to looping and excessive allocation, when the
computation should have just blocked on the black hole.  

Reported by Christian Höner zu Siederdissen <choener@tbi.univie.ac.at>
on glasgow-haskell-users.

The checkHeap function assumed the allocated part of the block contained only
alive objects and slops.  This was not true for blocks that are collected using
mark sweep.  The code in this patch skip the test for this kind of blocks.

This pulls from the upstream repos, for those packages which have
upstreams

Stops the docs going out of sync with the code.