This mainly affects non ASCII headers. MIME_ConvertCharset used to be used by message body conversion but now converters are cached in MIME object and this is not used. Currently, the main user is MIME decoder which needs conversion from MIME charset to UTF-8. This can be done by i18n util function and NS_ConvertUCS2toUTF8. Here is a data using Quantify. MIME decoder intl_decode_mime_part2_str() takes less time (about 57%) when not calling MIME_ConvertCharset(). intl_decode_mime_part2_str, calling MIME_ConvertCharset 40 36.72 21579.74 0.00 0.00 0.92 0.69 1.31 intl_decode_mime_part2_str, not calling MIME_ConvertCharset 40 37.60 12334.15 0.00 0.00 0.94 0.71 1.33

Forgot the labels of the last data. that is, # of callls Function time F+D time F time F+D time (% of Focus) Avg F time Min F time Max F time

Comment on attachment 62053 [details] [diff] [review] Removed MIME_ConvertCharset and related code. Looks good. R=ducarroz

Comment on attachment 62053 [details] [diff] [review] Removed MIME_ConvertCharset and related code. sr=sspitzer

checked in (on 12/18)

The following mimemoz2.cpp change commited in this bug removes the "res =" in the "else" branch, which would cause *uninitialized value* to be compared with 0 in the case of NULL encoder or decoder... @@ -764,16 +764,10 @@ PRInt32 res; char *convertedString = NULL; PRInt32 convertedStringLen; - if (!input_autodetect && encoder && decoder) + if (encoder && decoder) { res = ConvertUsingEncoderAndDecoder(input_line, input_length, encoder, decoder, &convertedString); convertedStringLen = (convertedString) ? nsCRT::strlen(convertedString) : 0; - } - else - { - // Now do conversion to UTF-8 for output - res = MIME_ConvertCharset(input_autodetect, input_charset, "UTF-8", input_line, input_length, - &convertedString, &convertedStringLen, NULL); } if (res != 0) {

Comment on attachment 64205 [details] [diff] [review] Changed mime_convert_charset to initialized the result value. R=ducarroz

Comment on attachment 64205 [details] [diff] [review] Changed mime_convert_charset to initialized the result value. sr=sspitzer

checked in

Naoki, could you help verifying this? Thanks a lot.

verified by LXR