Bare variable names (symbols) are required to be in the native encoding. On systems incapable of representing ñ in the native encoding (LC_ALL=C, or, e.g., KOI8-R), there is no way to preserve an ñ in a variable name.

On non-UTF-8 systems that can represent ñ in the native encoding, the code will work fine:

$ LC_ALL=en_GB.ISO-8859-15 luit R -q -s -e 'as.name("\uf1"); parse(text = "DT[, .N, a\U00F1o]$N[1L]")'
ñ
expression(DT[, .N, año]$N[1L])

If there is no ñ in the current locale, translateChar() internally called by parse() substitutes some text and you get a syntax error, but iconv seems to help:

# this works
LC_ALL=en_GB.ISO-8859-15 luit R -q -s -e 'text <- iconv("DT[, .N, a\U00F1o]$N[1L]", "UTF-8", ""); if (!is.na(text)) parse(text = text)'
# expression(DT[, .N, año]$N[1L])

# this doesn't crash
LC_ALL=C R -q -s -e 'text <- iconv("DT[, .N, a\U00F1o]$N[1L]", "UTF-8", ""); if (!is.na(text)) parse(text = text)'

Thanks @aitap. What's luit?

iconv() looks as good a solution as any -- definitely good to still run those tests on non-UTF-8 systems, rather than just skip if parsing fails.

luit converts between the UTF-8 terminal session and the non-UTF-8 encoding used by its child process.

#6559 demonstrates that we cannot rely on iconv() to return NA if conversion fails: on FreeBSD we instead get a?o.