They are decoded for graceful handling. &ampfoo is still a parse error.

Dear maintainers,

It is quite possible and feasible to have a more flexible
'html_entity_decode' that handles properly the named char
references that, for historical reasons, are allowed to
not be terminated with semicolon [1].

To sustain my claim, I invite you to examine Html-Cref
[2] -- a project that I developed quite recently which
implements several named character reference *parsers*
based on tries instead of hash tables.

Upon bringing into Html-Cref's framework PHP's function
'resolve_named_entity_html' and an adapted hash table
'ent_ht_html5' (all these as per the patch file enclosed;
the size of 'ent_ht_html5' was preserved), the standalone
binary obtained 'html-cref' is about 19% bigger then the
one built with the 'etrie' parser library: 203K vs. 163K.
Upon measurements, the new function 'html_cref_php_parse'
in 'src/html_cref_php.c' runs 4% slower than the fastest
trie-based parser on a 64-bit Intel Core I5-3210M machine.

Sincerely,

Stefan Vargyas.

[1] 12.2 Parsing HTML documents:
    12.2.5.73 Named character reference state
    https://html.spec.whatwg.org/#named-character-reference-state

[2] Html-Cref: Fast HTML Character References Decoder
    https://github.com/stvar/html-cref

The patch file referred by my comment above can be found at https://gist.github.com/stvar/df320f55d83cedac9fd7261256d20906.

Erratum to my initial comment:
203K is ~25% bigger than 163K.
Sorry for this.