https://bugs.documentfoundation.org/show_bug.cgi?id=172905

--- Comment #14 from Julien Nabet <[email protected]> ---
Digging a bit more, obviously the way of detection is different between
character sets but those like iso-8859-... start with a confidence of 0 and ICU
tries to find "prooves" it corresponds to one of them whereas character sets
like UTF-16LE start confidence from 10 and then it's adjusted.

Since numbers and punctuations are ignored to detect iso-8859, the rest of the
file has too few letters (accentuated or not) to increase confidence for
iso-8859.

Badfully, it means that a lot of statistics files (at least those containing °
encoded with B0 so some weather stats for example or any physics stats which
include temperature not encoded in UTF-8) may be wrongly detected with Chinese
glyphs.
So I still think the patch quoted in my previous comment (
+    if (ucsdet_getConfidence(match, &uerr) <= 10)
+        return;
)
may be relevant.

We could also add a section in option to enable/disable character sets detected
by default.
It would require some UI change + use of lib icu function "ucsdet_detectAll" to
filter out the non desired character sets.
(I've submitted a patch to display character sets detected by ICU and their
related confidence,
https://gerrit.libreoffice.org/c/core/+/208899/1/tools/source/stream/stream.cxx#824)

-- 
You are receiving this mail because:
You are the assignee for the bug.

Reply via email to