Truncating file names with Unicode characters

# Truncating file names with Unicode characters

When shortening file names that contain Unicode characters, git performs truncation without awareness of two-byte characters. That often leads to splitting a character in half and displaying a garbage byte that's left.

Unawareness of Unicode also means that filename length is calculated incorrectly and some output gets misaligned.

I have tested this with git 2.14.1 on Windows and with git 2.11.0 on Linux. My configuration includes setting `core.quotepath = off` to display Unicode paths.

# Example: `git log --stat`

## Bad output: half-characters and wrong text alignment

The last file name gets truncated in the middle of the character (`ˆ` is what's left of it). Text alignment is off because string lengths are calculated in bytes instead of characters.

    Extension/README.md                                |  28 +++++++++
    .../Catalog.Номенклатура.xml           |  32 ++++++++++
    .../Configuration.xml                              |   5 +-
    ...етПереработчика.ObjectModule.txt |  39 ++++++++++++
    ...cument.ОтчетПереработчика.xml |  68 +++++++++++++++++++++
    .../Enum.СтавкиНДС.xml                    |  24 ++++++++
    ...ˆирениеERPПотяркин_2018-06-05.cfe | Bin 0 -> 22018 bytes
    7 files changed, 195 insertions(+), 1 deletion(-)

## Good output with ASCII file names

Truncation and alignment are done right because each character is represented by a single byte.

    .../index.html                                             | 14
++++++++++++++
    docs/posts/2017/loops-in-power-query-m-language/index.html | 14
++++++++++++++
    .../index.html                                             |  7 +++++++
    .../temporary-virtual-environment-for-python/index.html    | 14
++++++++++++++
    .../index.html                                             | 14
++++++++++++++
    docs/posts/2018/getting-started-with-libpq/index.html      | 14
++++++++++++++
    .../index.html                                             | 14
++++++++++++++
    .../2018/unit-testing-in-power-query-m-language/index.html |  7 +++++++
    8 files changed, 98 insertions(+)

On Fri, Jun 08, 2018 at 09:25:29PM +0300, Vitaly Potyarkin wrote:

Yeah, I agree it would be nice for this to be more aware of multi-byte glyphs.

Unfortunately, we don't actually know what encoding the paths are in (and they might not even all be in the same encoding). But hopefully we could at least make the common case work a bit better. If we know that core.quotepath is off, then the user is expecting paths in their terminal encoding. So possibly we could just rely on i18n.logOutputEncoding at that point (which of course defaults to utf8).

If anyone is interested in working on it, the relevant code is in diff.c:show_stats(). Looks like we do a straight strlen() to get the width, which should become utf8_strwidth(). That would solve the problem where we use the wrong width.

To deal with the broken code points, I think you'd need to copy the name into a strbuf and use strbuf_utf8_replace() to truncate (it doesn't seem to be well documented, but you can grep for a few other callers).

-Peff