]> git.ipfire.org Git - thirdparty/binutils-gdb.git/commit - include/ChangeLog
libctf, ld: prohibit getting the size or alignment of forwards
authorNick Alcock <nick.alcock@oracle.com>
Tue, 5 Jan 2021 13:25:56 +0000 (13:25 +0000)
committerNick Alcock <nick.alcock@oracle.com>
Tue, 5 Jan 2021 14:53:39 +0000 (14:53 +0000)
commitffeece6ac2d4e4d2bf28c910b51c74ffc881c963
treedbb61a523d3289f68a024ed1c84a4290d75b364b
parent91e7ce2fd7b82f3982b453d1c6f4576621bc1c2d
libctf, ld: prohibit getting the size or alignment of forwards

C allows you to do only a very few things with entities of incomplete
type (as opposed to pointers to them): make pointers to them and give
them cv-quals, roughly. In particular you can't sizeof them and you
can't get their alignment.

We cannot impose all the requirements the standard imposes on CTF users,
because the deduplicator can transform any structure type into a forward
for the purposes of breaking cycles: so CTF type graphs can easily
contain things like arrays of forward type (if you want to figure out
their size or alignment, you need to chase down the types this forward
might be a forward to in child TU dicts: we will soon add API functions
to make doing this much easier).

Nonetheless, it is still meaningless to ask for the size or alignment of
forwards: but libctf didn't prohibit this and returned nonsense from
internal implementation details when you asked (it returned the kind of
the pointed-to type as both the size and alignment, because forwards
reuse ctt_type as a type kind, and ctt_type and ctt_size overlap).  So
introduce a new error, ECTF_INCOMPLETE, which is returned when you try
to get the size or alignment of forwards: we also return it when you try
to do things that require libctf itself to get the size or alignment of
a forward, notably using a forward as an array index type (which C
should never do in any case) or adding forwards to structures without
specifying their offset explicitly.

The dumper will not emit size or alignment info for forwards any more.

(This should not be an API break since ctf_type_size and ctf_type_align
could both return errors before now: any code that isn't expecting error
returns is already potentially broken.)

include/ChangeLog
2021-01-05  Nick Alcock  <nick.alcock@oracle.com>

* ctf-api.h (ECTF_INCOMPLETE): New.
(ECTF_NERR): Adjust.

ld/ChangeLog
2021-01-05  Nick Alcock  <nick.alcock@oracle.com>

* testsuite/ld-ctf/conflicting-cycle-1.parent.d: Adjust for dumper
changes.
* testsuite/ld-ctf/cross-tu-cyclic-conflicting.d: Likewise.
* testsuite/ld-ctf/forward.c: New test...
* testsuite/ld-ctf/forward.d: ... and results.

libctf/ChangeLog
2021-01-05  Nick Alcock  <nick.alcock@oracle.com>

* ctf-types.c (ctf_type_resolve): Improve comment.
(ctf_type_size): Yield ECTF_INCOMPLETE when applied to forwards.
Emit errors into the right dict.
(ctf_type_align): Likewise.
* ctf-create.c (ctf_add_member_offset): Yield ECTF_INCOMPLETE
when adding a member without explicit offset when this member, or
the previous member, is incomplete.
* ctf-dump.c (ctf_dump_format_type): Do not try to print the size of
forwards.
(ctf_dump_member): Do not try to print their alignment.
include/ChangeLog
include/ctf-api.h
ld/ChangeLog
ld/testsuite/ld-ctf/conflicting-cycle-1.parent.d
ld/testsuite/ld-ctf/cross-tu-cyclic-conflicting.d
ld/testsuite/ld-ctf/forward.c [new file with mode: 0644]
ld/testsuite/ld-ctf/forward.d [new file with mode: 0644]
libctf/ChangeLog
libctf/ctf-create.c
libctf/ctf-dump.c
libctf/ctf-types.c