From @Skarsnik

While writing a binding for Gumbo (A c99 lib to parse html) I came across an error on amd64 bits. One of the mapped struct has his size evaluated wrongly by nativesizeof (different than a call from sizeof see​: http://pastebin.com/FVagmG6T). It provokes random segfault because of that. It seem NC try to align the struct and fail. (8 bytes are added)

After isolating on a test file the struct (see attached file) and tracking some value on
https://github.com/MoarVM/MoarVM/blob/master/src/6model/reprs/CStruct.c#L110

I appear that after evaluating the size of the 2 last attributes (12 bytes each one)
https://github.com/MoarVM/MoarVM/blob/master/src/6model/reprs/CStruct.c#L256
realign the size and adding 2 time 4 bytes, ending with the wrong size of 104 bytes instead of 96.
I don't know if the align for the struct that is 12 bytes sized should be 4 bytes (and not ALIGNOF(void*) that is 8 for 64bits) or just the pointed (L256) adjustment is wrong

From @Skarsnik

testwrongsize.p6

From @Skarsnik

Le Mer 18 Nov 2015 07​:53​:46, scolinet@​gmail.com a écrit :

While writing a binding for Gumbo (A c99 lib to parse html) I came
across an error on amd64 bits. One of the mapped struct has his size
evaluated wrongly by nativesizeof (different than a call from sizeof
see​: http://pastebin.com/FVagmG6T). It provokes random segfault
because of that. It seem NC try to align the struct and fail. (8 bytes
are added)

After isolating on a test file the struct (see attached file) and
tracking some value on
https://github.com/MoarVM/MoarVM/blob/master/src/6model/reprs/CStruct.c#L110

I appear that after evaluating the size of the 2 last attributes (12
bytes each one)
https://github.com/MoarVM/MoarVM/blob/master/src/6model/reprs/CStruct.c#L256
realign the size and adding 2 time 4 bytes, ending with the wrong size
of 104 bytes instead of 96.
I don't know if the align for the struct that is 12 bytes sized should
be 4 bytes (and not ALIGNOF(void*) that is 8 for 64bits) or just the
pointed (L256) adjustment is wrong

The pastbin link is http://pastebin.com/FVagmG6T the ) get in the link

From @FROGGS

I can confirm that the alignment is wrong.

To prove, I created this repo​: https://github.com/FROGGS/p6-Gumbo
Running `perl6 t` gives​:
~/dev/gumbo$ perl6 t/00-structure-sizes.t
1..19
ok 1 - sizeof(size_t)
ok 2 - sizeof(GumboStringPiece)
ok 3 - sizeof(GumboSourcePosition)
ok 4 - sizeof(GumboVector)
ok 5 - sizeof(GumboTag)
ok 6 - sizeof(GumboNamespaceEnum)
ok 7 - sizeof(MyGumboTest)
not ok 8 - sizeof(GumboElement)

# Failed test 'sizeof(GumboElement)'
# at t/00-structure-sizes.t line 85
# expected​: '96'
# got​: '104'

Which hints at the last two attributes of the GumboElement structure, as the ticket already pointed out.

The RT System itself - Status changed from 'new' to 'open'

From @FROGGS

The fix for this misalignment is​: MoarVM/MoarVM@7b382acd61
Test got added here​: rakudo/rakudo@deeec10a2e

Closing ticket as resolved.

@FROGGS - Status changed from 'open' to 'resolved'