GSOC, Git Blog 3 - ZheNing Hu

Week3: Meticulousness is very important

What happened this week

I found a git cat-file bug this week, see:

git cat-file --batch=batman --batch-all-objects batman
fatal: object 00345b5fe0aa8f45e9d1289ceb299f0489ea3fe1 changed type!?

It seems that Git died for a strange reason: the type of an object changed? Is my Git object damaged? (Just like a friend of mine, use find . -type f -print0 | xargs -0 sed -i "s/\r//g" remove all the ‘\r’ of all files in a Git repository, this caused most of his Git commands to fail.) So I tested it under different linux platforms, they all have this same damage.

After a series of testing and debugging, I found that oid_object_info_extended() did not get the type of object.

So I submitted the patch for fix this bug to the Git mailing list in [PATCH] [GSOC] cat-file: fix —batch report changed-type bug, Peff tell us the essential reason for this bug:

In 845de33a5b (cat-file: avoid noop calls to sha1_object_info_extended, 2016-05-18), this patches skips the call to oid_object_info_extended() entirely when --batch-all-objects is in use, and the custom format does not include any placeholders that require calling it. The correct solution is to put checking if object_info is empty after setting object_info.typep.

Finally, thanks to the help of Jeff, I summit final patch in [PATCH v2 1/2] [GSOC] cat-file: handle trivial —batch format with —batch-all-objects.

Talk of experience as a story: Even experienced programmers make small mistakes, writing any code requires very careful thinking.

The checkpoints for rejecting %(raw) and --<lang> are incorrect. At Junio’s reminder, I changed the checkpoint from parse_ref_filter_atom() to verify_ref_format(). My mentor Christian pointed out some grammatical errors in the cover letter.
At the suggestion of Junio, I rebased 0efed94 ([GSOC] ref-filter: add %(raw) atom) on 1197f1a (ref-filter: introduce enum atom_type), they have a clash, after resolving the conflict, it’s better for me to consider the code I implemented before and the code I wrote now at the same time, I can find more problems and find better solutions.
I submitted the patch about %(rest), %(raw:textconv) and %(raw:filters) for ref-filter, they are used to simulate some functions of git cat-file, my mentor Hariom noticed one of the formatting issues, I am waiting for more reviews for the time being.

What’s the next step

As long as new atoms %(rest), %(raw:textconv) and %(raw:filters) for ref-filter can be accepted by Git, We can start to let cat-file use ref-filter logic on a large scale! Exciting! But the performance of ref-filter is still not good. Perhaps I need to find a new breakthrough in the performance bottleneck of ref-filter.

Thanks!

ZheNing Hu

第三周：细致非常重要

本周发生了什么

我这周发现了一个 git cat-file 的 bug，如下：

git cat-file --batch=batman --batch-all-objects batman
fatal: object 00345b5fe0aa8f45e9d1289ceb299f0489ea3fe1 changed type!?

Git 似乎因为一个很奇怪的原因挂掉了：对象的类型改变了？我的 Git 对象损坏了吗？（就像我的一个朋友，用 find . -type f -print0 | xargs -0 sed -i "s/\r//g" 去掉了 Git 仓库中所有文件的 ‘\r’，这导致他大部分 Git 命令都失效了。）于是我在不同的 Linux 平台上做了测试，它们都存在同样的“损坏”。

经过一系列的测试和调试，我发现 oid_object_info_extended() 没有获取到对象的类型。

因此我在 [PATCH] [GSOC] cat-file: fix —batch report changed-type bug 中向 Git 邮件列表提交了修复这个 bug 的补丁，Peff 告诉了我们这个 bug 的根本原因：

在 845de33a5b (cat-file: avoid noop calls to sha1_object_info_extended, 2016-05-18) 中，这个补丁在使用 --batch-all-objects 且自定义格式不包含任何需要调用它的占位符时，完全跳过了对 oid_object_info_extended() 的调用。正确的解决方案是在设置 object_info.typep 之后，再检查 object_info 是否为空。

最后，在 Jeff 的帮助下，我在 [PATCH v2 1/2] [GSOC] cat-file: handle trivial —batch format with —batch-all-objects 中提交了最终补丁。

作为经验之谈：即使是经验丰富的程序员也会犯小错误，编写任何代码都需要非常仔细地思考。

拒绝 %(raw) 和 --<lang> 的检查点不正确。在 Junio 的提醒下，我把检查点从 parse_ref_filter_atom() 改到了 verify_ref_format()。我的导师 Christian 指出了附信中的一些语法错误。
在 Junio 的建议下，我将 0efed94 ([GSOC] ref-filter: add %(raw) atom) 变基到了 1197f1a (ref-filter: introduce enum atom_type) 上，它们有冲突，解决冲突后，我可以同时考虑之前实现的代码和现在写的代码，这样能发现更多问题并找到更好的解决方案。
我提交了关于 ref-filter 的 %(rest)、%(raw:textconv) 和 %(raw:filters) 的补丁，它们用于模拟 git cat-file 的一些功能，我的导师 Hariom 注意到了一个格式问题，我暂时在等待更多审阅。

下一步计划

只要 ref-filter 的新原子 %(rest)、%(raw:textconv) 和 %(raw:filters) 能被 Git 接受，我们就可以开始大规模地让 cat-file 使用 ref-filter 逻辑了！令人兴奋！但 ref-filter 的性能仍然不够好。也许我需要在 ref-filter 的性能瓶颈上寻找新的突破。

谢谢！

ZheNing Hu