From: Martin Duerst <duerst@...>
Date: 2008-12-25T15:02:45+09:00
Subject: [ruby-core:20862] Re: [ruby-dev:37591] [BUG:trunk] [m17n] TestCSVFeatures fails  because of r20905

[Redirected to ruby-core so that James can also read this.]

Hello James,

This is an error report from Yugui about a csv test
failing on a Mac.

The reason for the failure is line 498 in lib/csv.rb,
in method CSV#inspect. This line reads:

str.map { |s| s.encode("ASCII-8BIT") }.join

The reason for the failure is that currently, filenames on a Mac
are labeled as being in an "encoding" of UTF8-MAC. The label
UTF8-MAC is used to mark the assumption that this string is in a
character normalization form particular to the Mac (mostly NFD,
but not for Korean, and not for CJKV compatibility ideographs,
as far as I understand).

There is in general no knowledge about character normalization with
respect to strings labeled UTF-8 (and even for UTF8-MAC, there is
no guarantee about character normalization at all). In my personal
view, the value of UTF8-MAC is questionable at least at the current
point in time where we do not handle character normalization in
any particular way. But for the current bug, that's actually a side
issue. We might be able to fix this by introducing a (dummy) conversion
from UTF8-MAC to UTF-8, but that won't actually fix the real problem.

The real problem is that the line above ignores that conversion
to ASCII-8BIT only works for US-ASCII characters, but not for
all the other characters that might appear e.g. in a filename.
The easiest fix for this, which is probably what was intended,
is to change the above line to

str.map { |s| s.force_encoding("ASCII-8BIT") }.join

A slightly more "user-friendly" fix is to change this to
something like:

begin
  str.join
rescue
  str.map { |s| s.force_encoding("ASCII-8BIT") }.join
end

This will only do a force_encoding if the encodings can't
be joined as is.

[The code above hasn't been tested; I don't have access to a Mac.]

Hope this helps.  Regards,    Martin.



At 12:27 08/12/25, Yugui (Yuki Sonoda) wrote:
>Yuguiです。
>
>r20905で、csvのテストが失敗するようになりました。pathのエンコーディング
>がUTF8-MACに変わったため、CSV#inspect内でString#encodeに失敗しているよう
>です。
>これはどう扱ったらよいものでしょうか。
>
>% make test-all TESTS=csv
>./miniruby -I../../lib -I.ext/common -I./- -r../../ext/purelib.rb
>../../runruby.rb --extout=.ext  -- "../../test/runner.rb" csv
>nil
>Loaded suite ../../test/runner
>Started
>..........E...........................................................................................................................
>Finished in 0.768623 seconds.
>
>  1) Error:
>test_inspect_is_smart_about_io_types(TestCSVFeatures):
>Encoding::ConverterNotFoundError: code converter not found (UTF8-MAC to
>ASCII-8BIT)
>    /Users/yugui/src/ruby/mri/test/csv/test_features.rb:233:in `block in
>test_inspect_is_smart_about_io_types'
>    /Users/yugui/src/ruby/mri/test/csv/test_features.rb:233:in
>`test_inspect_is_smart_about_io_types'
>
>134 tests, 1886 assertions, 0 failures, 1 errors, 0 skips
>make: *** [test-all] Error 1
>
>-- 
>Yugui <yugui@yugui.jp>
>http://yugui.jp
>私は私をDumpする


#-#-#  Martin J. Du"rst, Assoc. Professor, Aoyama Gakuin University
#-#-#  http://www.sw.it.aoyama.ac.jp       mailto:duerst@it.aoyama.ac.jp