Use diff transcoding
Bitbucket Data Center and Server supports transcoding for diffs. This allows it to convert files in encodings like EUC-JP, GB18030 and UTF-16 to UTF-8, so they are processed correctly by
git diff, which only supports UTF-8. Similar transcoding has been applied to the Bitbucket source view since it was released, so this change brings the diff view in line with the source view. Diff transcoding is applied to commit and pull request diffs, as well as the diff-to-previous view.
Git for Windows (旧称 msysgit) には、Unicode パスに関する既知の問題があります。差分トランスコーディングはサポートされているすべてのバージョンの Git for Windows で機能しますが、Unicode パスをサポートするには 1.8.0 以降が必要です。
差分トランスコーディングは (常に実行されるソース ビューのトランスコーディングとは異なり)、各リポジトリで明示的に有効化する必要があります。
There's a performance consideration with transcoding. It is implemented using Git's
textconv support, so using it adds overhead to displaying diffs. Where possible, the best approach, given
git only supports UTF-8 content, is to use UTF-8 encoding so that transcoding is not necessary. In repositories without non-UTF-8 content, diff transcoding should be left disabled. Other encodings are often a necessity, however, and for repositories containing such content enabling diff transcoding allows using the full range of Bitbucket features.
When transcoding is enabled,
git diff writes the before and after blobs to temporary files and invokes the
textconv script once for each file. The script Bitbucket installs uses Perl to send a request back to Bitbucket with the path to each temporary file. Bitbucket then opens each file, detects the encoding using the same algorithm the source view uses, converts the file to UTF-8 and streams it out for
git diff to use. After
git diff has invoked the
textconv script the temporary files it created are deleted.
Writing the blobs to disk, starting Perl and calling back into Bitbucket are all overhead processing compared to performing a diff without transcoding. How much overhead that is varies by the size of the diff. When nominally-sized files containing two or three thousand lines or less are being compared the overhead is miniscule, under 50 milliseconds on an average server. However, when comparing larger files the overhead can result in a noticeable delay displaying the diff.