Bug 295171 – Padding bits defeat intended HTMLEntityTableEntry size optimization

• Status: ASSIGNED
• Component: DOM

In https://github.com/WebKit/WebKit/pull/11089, HTMLEntityTableEntry was converted to a bit-field in order to decrease the size of kStaticEntityTable. HTMLEntityTableEntry theoretically uses 57 bits, but, due to padding bits, the size of HTMLEntityTableEntry is actually 12 bytes instead of 8 bytes (the same size it was before the bit-field change).

Test code:

struct HTMLEntityTableEntry {
    unsigned firstCharacter : 21;
    unsigned optionalSecondCharacter : 16;
    unsigned nameCharactersOffset : 14;
    unsigned nameLengthExcludingSemicolon : 5;
    unsigned nameIncludesTrailingSemicolon : 1;
};
static_assert(sizeof(HTMLEntityTableEntry) == 8);

Clang 20.1.3:

HTMLEntityTableEntry.cpp:8:15: error: static assertion failed due to requirement 'sizeof(HTMLEntityTableEntry) == 8'
    8 | static_assert(sizeof(HTMLEntityTableEntry) == 8);
      |               ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
HTMLEntityTableEntry.cpp:8:44: note: expression evaluates to '12 == 8'
    8 | static_assert(sizeof(HTMLEntityTableEntry) == 8);
      |               ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~
1 error generated.

GCC 11.4.0:

HTMLEntityTableEntry.cpp:8:44: error: static assertion failed
    8 | static_assert(sizeof(HTMLEntityTableEntry) == 8);
      |               ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~

(for context, this was found while I was writing https://www.ryanliptak.com/blog/better-named-character-reference-tokenization/)

[Darin Adler]

I tried moving optionalSecondCharacter to be the last member of the HTMLEntityTableEntry structure and that worked; just have to revise the code that builds the table to write out the members in the right order.

[Radar WebKit Bug Importer]

<rdar://problem/155138028>