Re: Added support for file list in single file update

[dhruva]

On Mon, Aug 25, 2014 at 9:16 PM, Shigio YAMAGUCHI <address@hidden> wrote:

2014-08-26 9:24 GMT+09:00 dhruva <address@hidden>:

The addition/changed/deleted is determined by the output of stat(). To keep it simple, we just maintain a list of files that have changed. It can be further optimized by using a prefix or separate file for added/deleted/changed file list but will complicate the code for minimal performance gain. I had tried it using the prefix approach, '+' for new files, '-' for deleted files and no prefix for modified/changed files.

 

Why do you withdraw the optimization? You said that stat() on a network

file system should be avoided. There was no person who opposed.

Since this ML is a place of arguments, nobody has duty to realize what

he asserted.

I cannot approve of provisional specification.

I believe that the problems in the prefix approach were solved by

the three options: --update-add, --update-delete, --update-change.

Generating a list of changes and putting them in 3 separate files is an overhead. Look from a usability perspective, we will have to find all added, deleted and modified files and put them in 3 different files.

In the current approach (my patch), we just put all the 3 types into the same file and it is easy to generate that file. Since I use perforce (p4) SCM at work, I just do "p4 opened|awk "STRIP_SOME_DEPOT_PATH" '{print $2}' | gtags -D -i -f -"

Overall, I would give priority to usability over slight improvement in performance. Doing it in batch mode gives use sufficient performance boost with no impact on usability.

with best regards,

dhruva