On Wed, Jul 17, 2019 at 10:49 PM Alex Kodat <alexko...@gmail.com> wrote:
> In Reland "[runtime] Speed up String::IsOneByte", the following check was > added to NonOneByteStart in src/objects/string.h: > > DCHECK(IsAligned(reinterpret_cast<Address>(chars), sizeof(uc16))); > > This is part of the very clever word at a time detection of code points > > 256. Unfortunately for us, this code ends up getting called for data we > pass to String::NewFromTwoByte. And while the name gives one pause, the > comment says > > /** Allocates a new string from UTF-16 data.*/ > > so we've been using this against UTF-16 data which, UTF-16 being a > serialization mechanism might not be two-byte aligned (ours is often in > protocol buffer data). In any case, String::NewFromTwoByte > calls Factory::NewStringFromTwoByte with the unaligned string which then > calls String::IsOneByte with the unaligned string which then fails with the > DCHECK. Without the DCHECK it works because the code that's scanning for > the start of word alignment works fine for non-two-byte aligned data so > this is only really an issue for debug builds though admittedly the > non-one-byte scan won't be nearly as efficient for non-aligned data so > maybe the code is doing us a favor by hitting us on the head. > > For non-two-byte aligned data, the clever algorithm could still be made to > work by flipping the endian-specific word mask and a bit of up-front > fiddling but far be it for me to advocate anyone going to that trouble > though I guess if I were told that such a change would be accepted i'd be > willing to contribute such a fix, myself. Another way to fix it would be to > create the string in the heap as a two-byte string and then scan the > aligned data in the heap, flipping the one-byte bit (or whatever indicates > one-byte) in the string object. But haven't researched whether there is a > call to do this bit flip. > > In any case, it doesn't seem nice that V8 no longer has an unaligned > UTF-16 to string conversion mechanism, at least not one that works for > debug builds. > > Opinions? > > -- > -- > v8-users mailing list > v8-users@googlegroups.com > http://groups.google.com/group/v8-users > --- > You received this message because you are subscribed to the Google Groups > "v8-users" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to v8-users+unsubscr...@googlegroups.com. > To view this discussion on the web visit > https://groups.google.com/d/msgid/v8-users/7f027ba5-d190-45a0-ad94-954e61ec8a4f%40googlegroups.com > <https://groups.google.com/d/msgid/v8-users/7f027ba5-d190-45a0-ad94-954e61ec8a4f%40googlegroups.com?utm_medium=email&utm_source=footer> > . > For more options, visit https://groups.google.com/d/optout. > -- -- v8-users mailing list v8-users@googlegroups.com http://groups.google.com/group/v8-users --- You received this message because you are subscribed to the Google Groups "v8-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to v8-users+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/v8-users/CAH3p7oPiZ2ZftpOUvOSaXSkvntbVQVoPaJHLekMWiam7yUYJcQ%40mail.gmail.com. For more options, visit https://groups.google.com/d/optout.
