On Tue, Jul 4, 2023 at 11:50 AM Thomas Schwinge <tho...@codesourcery.com> wrote:
>
> Hi!
>
> I came across this one here on my way working through another (somewhat
> related) GTY issue. I generally do understand the issue here, but do
> have a question about 'unsigned int len' field in
> 'libcpp/include/symtab.h:struct ht_identifier':
>
> On 2022-10-18T18:14:54-0400, Lewis Hyatt via Gcc-patches
> <gcc-patches@gcc.gnu.org> wrote:
> > When a GTY'ed struct is streamed to PCH, any plain char* pointers it
> > contains
> > (whether they live in GC-controlled memory or not) will be marked for PCH
> > output by the routine gt_pch_note_object in ggc-common.cc. This routine
> > special-cases plain char* strings, and in particular it uses strlen() to get
> > their length.
>
> Oh, wow, this special casing for strings... 8-|
>
> > Thus it does not handle strings with embedded null bytes, but it
> > is possible for something PCH cares about (such as a string literal token
> > in a
> > macro definition) to contain such embedded nulls. To fix that up, add a new
> > GTY option "string_length" so that gt_pch_note_object can be informed the
> > actual length it ought to use, and use it in the relevant libcpp structs
> > (cpp_string and ht_identifier) accordingly.
>
> For your test case:
>
> > --- /dev/null
> > +++ b/gcc/testsuite/g++.dg/pch/pch-string-nulls.C
> > @@ -0,0 +1,3 @@
> > +// { dg-do compile { target c++11 } }
> > +#include "pch-string-nulls.H"
> > +static_assert (X[4] == '[' && X[5] == '!' && X[6] == ']', "error");
>
> > --- /dev/null
> > +++ b/gcc/testsuite/g++.dg/pch/pch-string-nulls.Hs
> > @@ -0,0 +1,2 @@
> > +/* Note that there is a null byte following "ABC". */
> > +#define X R"(ABC^@[!])"
>
> ..., I understand how the following is necessary:
>
> > --- a/libcpp/include/cpplib.h
> > +++ b/libcpp/include/cpplib.h
> > @@ -179,7 +179,11 @@ enum c_lang {CLK_GNUC89 = 0, CLK_GNUC99, CLK_GNUC11,
> > CLK_GNUC17, CLK_GNUC2X,
> > /* Payload of a NUMBER, STRING, CHAR or COMMENT token. */
> > struct GTY(()) cpp_string {
> > unsigned int len;
> > - const unsigned char *text;
> > +
> > + /* TEXT is always null terminated (terminator not included in len); but
> > this
> > + GTY markup arranges that PCH streaming works properly even if there
> > is a
> > + null byte in the middle of the string. */
> > + const unsigned char * GTY((string_length ("1 + %h.len"))) text;
> > };
>
> (That is, the test case FAILs with that one reverted.)
>
> However, this one did confuse me:
>
> > --- a/libcpp/include/symtab.h
> > +++ b/libcpp/include/symtab.h
> > @@ -29,7 +29,10 @@ along with this program; see the file COPYING3. If not
> > see
> > typedef struct ht_identifier ht_identifier;
> > typedef struct ht_identifier *ht_identifier_ptr;
> > struct GTY(()) ht_identifier {
> > - const unsigned char *str;
> > + /* This GTY markup arranges that the null-terminated identifier would
> > still
> > + stream to PCH correctly, if a null byte were to make its way into an
> > + identifier somehow. */
> > + const unsigned char * GTY((string_length ("1 + %h.len"))) str;
> > unsigned int len;
> > unsigned int hash_value;
> > };
>
> I did wonder whether that's an actual or just a theoretical concern, to
> have 'ht_identifier's with embedded NULs? If an actual concern, can we
> get a test case constructed? Otherwise, should we revert this hunk,
> given that we have this auto-'strlen' handling, ignorant of embedded
> NULs, in a lot of other places?
>
> But then I realized that possibly we do maintain 'len' here not for
> correctness but as an optimization (trading an 'unsigned int' for
> repeated 'strlen' calls)? My quick testing with the attached
> "[RFC] Verify no embedded NULs in 'struct ht_identifier'" might confirm
> this theory: no regressions (..., but I didn't bootstrap, and ran only
> parts of the testsuite). (Not proposing that RFC for 'git push', of
> course.)
>
> If that's indeed the intention here, I shall change/add source code
> commentary to describe this rationale for this dedicated 'len' field
> (plus, that handling of embedded NULs falls out of that, automatically).
>
> For reference, this 'len' field has existed "forever". Before
> 'struct ht_identifier' was added in
> commit 2a967f3d3a45294640e155381ef549e0b8090ad4 (Subversion r42334), we
> had in 'gcc/cpplib.h:struct cpp_hashnode': 'unsigned short len', or
> earlier 'length', earlier in 'gcc/cpphash.h:struct hashnode':
> 'unsigned short length', earlier 'size_t length' with comment:
> "length of token, for quick comparison", erlier 'int length', ever since
> the 'gcc/cpp*' files were added in
> commit 7f2935c734c36f84ab62b20a04de465e19061333 (Subversion r9191).
>
I don't think there is currently any possibility for a null byte to
end up in an ht_identifier's string. I assumed that ht_identifier
stores the length as an optimization (especially since it doesn't take
up any extra space on 64-bit platforms, given the 32-bit hash code is
stored as well there.) I created the string_length GTY markup mainly
to support another patch that I have still pending review, which I
thought would increase the likelihood of PCH needing to handle null
bytes in general. When I did that, I added the markup to ht_identifier
simply because the length was already there, so there was no reason
not to add it. It does save a few cycles when streaming out the PCH,
but I doubt it is meaningful.
-Lewis
