Unicode→LaTeX export produces unbraced stacked accents for precomposed Vietnamese letters (`ễ`, `ắ`), which biber then half-decodes into invalid input

Debug log ID

2FC9N2GZ-refs-euc/✓/✓

What happened?

What happens

An author name stored in Zotero as plain Unicode:

Nguyễn, Xuân Thắng

is exported by Better BibTeX as:

author = {Nguy{\~\^e}n, Xu{\^a}n Th{\'\\\u a}ng and Tran, Huu Tam and ...}

Two characters are involved, both of which need two stacked accents:

In both cases the inner accent command is not braced: BBT emits \~\^e, not \~{\^e}.

Why it breaks downstream

Feeding that .bib to biber (3.19 / biber 2.19) and pdfLaTeX with \\\usepackage[utf8]{inputenc} and \\\usepackage[T1]{fontenc}:

1. biber's LaTeX→Unicode decoding resolves only the inner accent, and writes into main.bbl:

   复制

   family={Nguy{\~ê}n},
   given={Xuân\bibnamedelima Th{\'ă}ng},

   i.e. an accent command applied to an already-composed UTF-8 character.

2. inputenc makes the lead byte of ê (0xC3) an active character that reads the byte after it. \~ grabs that active character as its argument, so the trailing byte is left alone and pdfLaTeX stops:

   复制

   ./file.tex:83: LaTeX Error: Invalid UTF-8 byte sequence (^^c3\global).
   ./file.tex:83: LaTeX Error: Invalid UTF-8 byte "AA.

The braced form is not a workaround either: writing Nguy{\~{\^e}}n in the .bib by hand produces exactly the same main.bbl as the unbraced form, and the same failure. Only a .bib holding the precomposed characters themselves compiles, once \DeclareUnicodeCharacter declares them:

\DeclareUnicodeCharacter{1EC5}{\~{\^e}}
\DeclareUnicodeCharacter{1EAF}{\'{\\\u a}}

Minimal reproduction

t.bib — the three encodings, and what biber makes of each:

@article{a1, title={A}, author={Nguy{\~\^e}n, Xu{\^a}n Th{\'\\\u a}ng},   year=2015, journal={J}}
@article{a2, title={B}, author={Nguy{\~{\^e}}n, Xu{\^a}n Th{\'{\\\u a}}ng}, year=2015, journal={J}}
@article{a3, title={C}, author={Nguyễn, Xuân Thắng},                    year=2015, journal={J}}

a1 → family={Nguy{\~ê}n}   given={Xuân Th{\'ă}ng}    pdfLaTeX: error
a2 → family={Nguy{\~ê}n}   given={Xuân Th{\'ă}ng}    pdfLaTeX: error
a3 → family={Nguyễn}       given={Xuân Thắng}        pdfLaTeX: ok

What I would expect

Either of these would fix it, and the second looks more robust:

1. brace nested accent commands — {\~{\^e}}, {\'{\\\u a}} — which at least makes the exported LaTeX well-formed, though it does not survive the biber round-trip; or
2. leave precomposed characters that require stacked accents as Unicode in the export, the way a3 above works. They are the only form the whole chain accepts.

The same question applies to the rest of the Vietnamese block (U+1EA0–U+1EF9) and to any other script whose letters decompose into two or more accents.

Workaround in use

Strip the diacritics from the name in Zotero, which loses the correct spelling of a colleague's name.