Hi everyone,

I'm Tian Yuchen and I'm planning to apply for GSoC this year!

Instead of pasting a giant wall of text into this email, I have drafted my proposal in Google Doc. I thought it might be easier for everyone to leave inline comments and suggestions there. (Of course, if you're more accustomed to email replies, you can also quote the content from the doc in your response. Thank you.)

Here is the link:

https://docs.google.com/document/d/1t2sznOvnPz-9tOzVMH--pLxzRqYSJCFzqVWBVfL_NP8/edit?tab=t.0#heading=h.c3c40ftj1ilv

Feel free to provide feedback!

Regards,

Yuchen

On Sun, Feb 22, 2026 at 11:29 PM Tian Yuchen <a3205153416@gmail.com> wrote:

I believe that a giant wall of text is the appropriate way to send a proposal to the Git community. I will advise you to send that giant of text actually. It is easier for the community to review and give feedback. Also future gsoc participants can also learn from it. By telling the reviewer to go through the link to the docs and then copy it on the to the email just to reply is giving them an extra lot of work to do. So send it through text and make it easy for people to review.

Thank you.

On 2/23/26 02:34, Usman Akinyemi wrote:

Makes sense. I will format the proposal into plain text and send it as v2 shortly so it's easier to review inline and properly archived.

Thanks,

Yuchen

Hello everyone,

I'm Tian Yuchen and I'm planning to apply for GSoC project this year. I hope you can take the time to review my proposal.

Please feel free to leave feedback!

Google Docs link:

https://docs.google.com/document/d/1t2sznOvnPz-9tOzVMH--pLxzRqYSJCFzqVWBVfL_NP8/edit?tab=t.0#heading=h.c3c40ftj1ilv

Refactoring in order to reduce Git's global state =================================================

PERSONAL INFORMATION
--------------------
Name: Tian Yuchen
E-mail: a3205153416@gmail.com
Phone number: +65 98740318
Time-zone: UTC + 08:00
Github: https://github.com/malon7782

Education: NTU, Singapore
Year: Year 1 semester 2
Degree: Electrical and Electronic Engineering (EEE)

PRE GSOC -------- I have always held a deep passion for the open-source community. Although I wasn't a computer science major, I tinkered with open-source projects long before college. I have solid hands-on experience in C programming and system-level debugging.

I use Ubuntu 24.04 on a daily basis, so I am proficient in using the Linux command line and CLI tools.

I have contributed to the Git community by sending patches. Since my first commit (17/1/2026), I have maintained a nearly daily contribution. Here is the list of contributions I have made:

* [PATCH v1] t1005: modernize "! test -f" to "test_path_is_missing"
  
https://lore.kernel.org/git/20260117062515.319664-1-a3205153416@gmail.com/
   This patch is my microproject, the first contribution I made to the 
codebase.
   [Graduated to 'master']

* [PATCH v2] t2203: avoid masking exit codes in git status
  
https://lore.kernel.org/git/20260118043537.338769-1-a3205153416@gmail.com/#t

* [PATCH v2] symlinks: use unsigned int for flags
  
https://lore.kernel.org/git/20260120152219.398999-1-a3205153416@gmail.com/
   [Will merge to 'next']

* [PATCH v4] t/perf/p3400: speed up setup using fast-import
  
https://lore.kernel.org/git/20260130170123.642344-1-a3205153416@gmail.com/
   [Will merge to 'master']

* Re: [PATCH] [RFC] attr: use local repository state in read_attr
  
https://lore.kernel.org/git/cc2f400e-49c2-4de0-9c51-9a5c0294735e@gmail.com/
   Code review. To verify the performance loss, I wrote a test script to
   measure the time difference before and after the modification.

* Re: Bug: git add :!x . exits with error when x is in .gitignore
  
https://lore.kernel.org/git/1d560aa1-d452-47f5-aaf2-4cb1ccdab100@gmail.com/
   Code review. Pointed out logical error.

* [PATCH v10] setup: allow cwd/.git to be a symlink to a directory
  
https://lore.kernel.org/git/20260220164512.216901-1-a3205153416@gmail.com/
   In progress.
   After over half a month of discussions, repeated refactoring, and code
   reviews, I delved deep into setup.c. I gained insights into Git's 
design philosophy, and learned the art of striking a balance in 
developer communication. It took me a large amount of time and effort to 
thoroughly understand every line of the code. I often found myself 
poring over the call chain of a single function well into the night.... 
But I persevered until the end, and I believe my patience will see me 
through even larger projects.

ABOUT THE PROJECT -----------------

-- Synopsis

As far as I know, the Git community is actively working towards 'libification' - making Git's internal machinery reusable as a C library. The extensive reliance on global state is a major roadblock to this goal.

Many core functions implicitly read environment variables and store them in global static variables. This can cause several issues:

   1. Global variables prevent Git's core functions from being executed 
safely in multi-threaded contexts.
   2. When Git is called multiple times within the same process, global 
states can lead to memory leaks or incorrect behaviors.
   3. Unit testing becomes difficult because the environment must be 
artificially manipulated before calling functions.

Take a look at this example from environment.c:

     206 const char *get_commit_output_encoding(void)
     207 {
     208     return git_commit_encoding ? git_commit_encoding : "UTF-8";
     209 }

If Git is invoked as a C library by a multi-threaded server:
- Thread A formats a commit for Repo A (using GBK);
- Thread B concurrently formats a commit for Repo B (using UTF-8);

Then they will race to read and overwrite the exact same global `git_commit_encoding` pointer, which is not what we expect. Therefore, we have to refactor these environment variables by moving them from global scope into a well-defined and encapsulated context.

-- Approach

The task at hand can be summed up in one sentence: repackage the global variables into the `struct repository` structure. In other words:

     [ Current ]
     Core functions --------reads-------> Global variables (via getenv)
                                          [Thread unsafe]

     [ Target ]
     Core functions ----passes context--> struct repository
                                                 | owns
                                                 v
                                          struct git_env

Although the principle is simple, the scope of changes is extensive. The following three-step approach can serve as a guiding principle for it:

   1. Identify isolated environment variables currently residing in the
      global scope. Introduce a dedicated structure to hold these states,
      e.g. `struct git_env` within the `struct repository`.
   2. Modify the function signatures within the call chain to accept the
      context, e.g., `struct repository *repo`, instead of relying on
      implicit globals. External callers of the functions must be
      carefully audited to prevent regressions.
   3. Safely remove the old global variables and macro definitions. Tools
      such as AddressSanitizer can be helpful to ensure that the new
      struct-based lifecycle introduces zero memory leaks.

Additionally, given the anticipated high volume of commits, we must ensure each patch is independent and atomic, preventing any user-untraceable or unexplainable bugs from occurring in the codebase at any state.

AVAILABILITY ------------ Fortunately, my summer vacation coincides with the GSoC work period. I will treat this project as my primary focus, dedicating a minimum of 35 hours per week. If needed, I can work a 9-to-5 schedule.

I will have a significant head start to draft RFC patches before the official coding period even begins. Having this buffer period allows me to go through the rigorous code review process within the Git community with greater ease.

TIMELINE & MILESTONES --------------------- Considering the differences between this project and other projects on the idea list, rather than hoarding massive changes, I will submit 3-to-5-patch series frequently to respect reviewers' time and maintain a steady velocity.

Below is the tentative schedule I have prepared for myself:

* Community Bonding (May 1 - May 25): Planning & RFC
   - May 1 - May 7: Wrap up university finals. Discuss and finalize the
     prioritized list of subsystems with my mentor.
   - May 8 - May 25: Define the core context container. Draft and submit
     the initial RFC patch series for this new data structure.

* Phase 1 (May 26 - July 10): Foundation
   - Weeks 1-2: Plumb the context pointer (`struct repository *repo`) 
through call chains for simple variables (e.g., boolean flags or integer 
configs).
   - Weeks 3-4: Audit and update external callers to use the new API.
   - Weeks 5-6: Submit the first major refactoring patch series. Address
     mailing list feedback and resolve merge conflicts. (Midterm Evaluation)

* Phase 2 (July 11 - August 18): Complex Migration & Cleanup
   - Weeks 7-8: Refactor higher-complexity variables (e.g., path-related 
globals).
   - Weeks 9-10: Compile the codebase with AddressSanitizer and run the 
full test suite to execute strict memory leak checks.
   - Weeks 11-12: Remove unused global macro definitions and static 
variables. Update internal documentation and write the final GSoC report.

(The above is for reference only. Personally, I always finish tasks faster than planned ;)

~$ git checkout HEAD@{postGSoC} ------------------------------- This past month since joining the Git community has been the most enjoyable month of my programming journey. To quote a close friend of mine (who is applying for the Neovim GSoC project):

   "Only fools chase trends; open source is the game for the brave."

The words may be blunt, but the logic holds true. This statement surely resonates with me (and maybe many other GSoC contributors): our passion for code and open-source drives us forward.

Even if I didn't make the cut, so what? ~$ git reset --hard... Just kidding. The Git codebase is far too interesting to abandon now.

------------------------------------------------------------------------- Changes since V1:

  - Transfer the text from Google Docs to here.

Regards,

Yuchen

Hi mentors and the git community,

Here is the proposal V3.

Refactoring in order to reduce Git's global state =================================================

PERSONAL INFORMATION
--------------------
Name: Tian Yuchen
E-mail: a3205153416@gmail.com
Phone number: +65 98740318
Time-zone: UTC + 08:00
Github: https://github.com/malon7782

Education: NTU, Singapore
Year: Year 1 semester 2
Degree: Electrical and Electronic Engineering (EEE)

PRE GSOC -------- I have always held a deep passion for the open-source community. Although I wasn't a computer science major, I tinkered with open-source projects long before college. I have solid hands-on experience in C programming and system-level debugging.

I use Ubuntu 24.04 on a daily basis, so I am proficient in using the Linux command line and CLI tools.

I have contributed to the Git community by sending patches. Since my first commit (17/1/2026), I have maintained a nearly daily contribution. Here is the list of contributions I have made:

* [PATCH v1] t1005: modernize "! test -f" to "test_path_is_missing"

https://lore.kernel.org/git/20260117062515.319664-1-a3205153416@gmail.com/
   This patch is my microproject, the first contribution I made to the 
codebase.
   [Graduated to 'master']

* [PATCH v2] t2203: avoid masking exit codes in git status

https://lore.kernel.org/git/20260118043537.338769-1-a3205153416@gmail.com/#t

* [PATCH v2] symlinks: use unsigned int for flags

https://lore.kernel.org/git/20260120152219.398999-1-a3205153416@gmail.com/
   [Will merge to 'next']

* [PATCH v4] t/perf/p3400: speed up setup using fast-import

https://lore.kernel.org/git/20260130170123.642344-1-a3205153416@gmail.com/
   [Will merge to 'master']

* Re: [PATCH] [RFC] attr: use local repository state in read_attr

https://lore.kernel.org/git/cc2f400e-49c2-4de0-9c51-9a5c0294735e@gmail.com/
   Code review. To verify the performance loss, I wrote a test script to
   measure the time difference before and after the modification.

* Re: Bug: git add :!x . exits with error when x is in .gitignore

https://lore.kernel.org/git/1d560aa1-d452-47f5-aaf2-4cb1ccdab100@gmail.com/
   Code review. Pointed out logical error.

* [PATCH v10] setup: allow cwd/.git to be a symlink to a directory

https://lore.kernel.org/git/20260220164512.216901-1-a3205153416@gmail.com/
   [Under review]
   After over half a month of discussions, repeated refactoring, and code
   reviews, I delved deep into setup.c. I gained insights into Git's 
design philosophy, and learned the art of striking a balance in 
developer communication. It took me a large amount of time and effort to 
thoroughly understand every line of the code. I often found myself 
poring over the call chain of a single function well into the night.... 
But I persevered until the end, and I believe my patience will see me 
through even larger projects.

ABOUT THE PROJECT -----------------

-- Synopsis

As far as I know, the Git community is actively working towards 'libification' - making Git's internal machinery reusable as a C library. The extensive reliance on global state is a major roadblock to this goal.

Many core functions implicitly read environment variables and store them in global static variables. This can cause several issues:

   1. Global variables prevent Git's core functions from being executed 
safely in multi-threaded contexts. For example, When unexpected states 
(e.g., a permission denied error when probing a directory), they often 
rely on the global state to decide whether to call die(), which 
internally calls exit(). It’s fine for a standalone CLI tool, but for a 
linked C library used by a long-running multi-threaded server, a single 
die() call will kill the entire host process. Structured status, instead 
of fatal exits, should be returned.
   2. When Git is called multiple times within the same process, global 
states can lead to memory leaks or incorrect behaviors.
   3. Unit testing becomes difficult because the environment must be 
artificially manipulated before calling functions.

Take a look at this example from environment.c:

     206 const char *get_commit_output_encoding(void)
     207 {
     208     return git_commit_encoding ? git_commit_encoding : "UTF-8";
     209 }

If Git is invoked as a C library by a multi-threaded server:
- Thread A formats a commit for Repo A (using GBK);
- Thread B concurrently formats a commit for Repo B (using UTF-8);

Then they will race to read and overwrite the exact same global `git_commit_encoding` pointer, which is not what we expect. Therefore, we have to refactor these environment variables by moving them from global scope into a well-defined and encapsulated context.

-- Approach

The task at hand goes beyond simply repackaging the global variables into the struct repository structure. Based on my recent experience refactoring setup.c, I realized that libification requires careful management of variable lifecycles and api boundaries:

     [ Current ]
     Core functions --------reads-------> Global variables (via getenv)
                                          [Thread unsafe]

     [ Target ]
     Core functions ----passes context--> struct repository
                                                 | owns
                                                 v
                                          struct git_env

Although the principle is simple, the scope of changes is extensive. The following three-step approach can serve as a guiding principle for it:

   1. Identify isolated environment variables currently residing in the
      global scope. Introduce a dedicated structure to hold these states,
      e.g. `struct git_env` within the `struct repository`.
   2. Instead of blindly passing struct repository *repo down into every
      single low-level library function, bubbling the dependency up is
      the true goal. External callers of the functions must be carefully
      audited to prevent regressions.
   3. Safely remove the old global variables and macro definitions. Tools
      such as AddressSanitizer can be helpful to ensure that the new
      struct-based lifecycle introduces zero memory leaks.
   4. Many globals like are parsed once and remain available globally.
      New data flow might need to be designed to maintain the lazy-
      loading efficiency.

Additionally, given the anticipated high volume of commits, we must ensure each patch is independent and atomic, preventing any user-untraceable or unexplainable bugs from occurring in the codebase at any state.

AVAILABILITY ------------ Fortunately, my summer vacation coincides with the GSoC work period. I will treat this project as my primary focus, dedicating a minimum of 35 hours per week. If needed, I can work a 9-to-5 schedule.

I will have a significant head start to draft RFC patches before the official coding period even begins. Having this buffer period allows me to go through the rigorous code review process within the Git community with greater ease.

TIMELINE & MILESTONES --------------------- Considering the differences between this project and other projects on the idea list, rather than hoarding massive changes, I will submit 3-to-5-patch series frequently to respect reviewers' time and maintain a steady velocity.

Below is the tentative schedule I have prepared for myself:

* Community Bonding (May 1 - May 25): Planning & RFC
   - May 1 - May 7: Wrap up university finals. Discuss and finalize the
     prioritized list of subsystems with my mentor.
   - May 8 - May 25: Define the core context container. Draft and submit
     the initial RFC patch series for this new data structure.

* Phase 1 (May 26 - July 10): Foundation
   - Weeks 1-2: Plumb the context pointer (`struct repository *repo`) 
through call chains for simple variables (e.g., boolean flags or integer 
configs).
   - Weeks 3-4: Audit and update external callers to use the new API.
   - Weeks 5-6: Submit the first major refactoring patch series. Address
     mailing list feedback and resolve merge conflicts. (Midterm Evaluation)

* Phase 2 (July 11 - August 18): Complex Migration & Cleanup
   - Weeks 7-8: Refactor higher-complexity variables (e.g., path-related 
globals).
   - Weeks 9-10: Compile the codebase with AddressSanitizer and run the 
full test suite to execute strict memory leak checks.
   - Weeks 11-12: Remove unused global macro definitions and static 
variables. Update internal documentation and write the final GSoC report.

(The above is for reference only. Personally, I always finish tasks faster than planned 😉)

~$ git checkout HEAD@{postGSoC} ------------------------------- This past month since joining the Git community has been the most enjoyable month of my programming journey. To quote a close friend of mine (who is applying for the Neovim GSoC project):

   "Only fools chase trends; open source is the game for the brave."

The words may be blunt, but the logic holds true. This statement surely resonates with me (and maybe many other GSoC contributors): our passion for code and open-source drives us forward.

Even if I didn't make the cut, so what? ~$ git reset --hard... Just kidding. The Git codebase is far too interesting to abandon now.

------------------------------------------------------------------------- Changes since V3:

  - Based on reviewing last year's contributors' changes and recent 
experience modifying setup.c, additional descriptions have been added to 
the synopsis & approach section.

Tian Yuchen <a3205153416@gmail.com> writes:

Hello Tian,

[snip]

Well it depends, we already have `struct repo_settings`, and individual settings within the `struct repository` struct. It would be a very case by case basis, to understand which variables fit where.

Yes, we also have CI jobs for GitLab and GitHub which do this already, you can run them locally too, meson makes it very easy to do this too:

  $ meson setup address --fatal-meson-warnings --warnlevel 3 --werror \
  --wrap-mode nofallback -Dfuzzers=true -Db_sanitize=address \
  -Db_lundef=false
  $ cd address
  $ meson test

I think you're missing a reference in this sentence.

What is the 'core context container' here?

Thanks for the proposal :)

Hi karthik,

Thank you so much for the review.

That makes sense to me. I will update the approach to emphasize a case-to-case analysis & mapping variables to their rightful existing homes.

Thank you for providing the information above. I have integrated this into the V4 proposal. It indeed looks much more robust in terms of memory leak auditing and other checks.

Sorry that was a typo. While writing the proposal, I went back to the source code to confirm the function name here, but I forgot to add it back in LOL. I meant 'editor_program' :)

Emmmm...It was referring to the 'struct git_env' idea (or something like that), which is flawed as you mentioned earlier. Will revise the timeline: the bonding period will be spent categorizing the targeted global variables and determining their appropriate stuctural destinations (via RFC patches? I don't know if it's proper behavior).

Will incorporate all these refinements and send out V4 within ~3 days (I have midterm tests these days). Thanks again for your time and patience ;)

Regards,

Yuchen

Hi karthink,

One thing I forgot to mention—I plan to practice with a simple target following the proposal's process to see if it works in the coming days. If you have time to review it and offer some feedback, I'd be most grateful!

Thanks,

Yuchen

Hi all,

Here is the V4 patch.

Thanks Karthink Nayak <karthink.188@gamil.com>, for help and guidance.

Refactoring in order to reduce Git's global state =================================================

PERSONAL INFORMATION
--------------------
Name: Tian Yuchen
E-mail: a3205153416@gmail.com
Phone number: +65 98740318
Time-zone: UTC + 08:00
Github: https://github.com/malon7782

Education: NTU, Singapore
Year: Year 1 semester 2
Degree: Electrical and Electronic Engineering (EEE)

PRE GSOC -------- I have always held a deep passion for the open-source community. Although I wasn't a computer science major, I tinkered with open-source projects long before college. I have solid hands-on experience in C programming and system-level debugging.

I use Ubuntu 24.04 on a daily basis, so I am proficient in using the Linux command line and CLI tools.

I have contributed to the Git community by sending patches. Since my first commit (17/1/2026), I have maintained a nearly daily contribution. Here is the list of contributions I have made:

* [PATCH v1] t1005: modernize "! test -f" to "test_path_is_missing"

https://lore.kernel.org/git/20260117062515.319664-1-a3205153416@gmail.com/
   This patch is my microproject, the first contribution I made to the 
codebase.
   [Graduated to 'master']

* [PATCH v2] t2203: avoid masking exit codes in git status

https://lore.kernel.org/git/20260118043537.338769-1-a3205153416@gmail.com/#t

* [PATCH v2] symlinks: use unsigned int for flags

https://lore.kernel.org/git/20260120152219.398999-1-a3205153416@gmail.com/
   [Will merge to 'next']

* [PATCH v4] t/perf/p3400: speed up setup using fast-import

https://lore.kernel.org/git/20260130170123.642344-1-a3205153416@gmail.com/
   [Will merge to 'master']

* Re: [PATCH] [RFC] attr: use local repository state in read_attr

https://lore.kernel.org/git/cc2f400e-49c2-4de0-9c51-9a5c0294735e@gmail.com/
   Code review. To verify the performance loss, I wrote a test script to
   measure the time difference before and after the modification.

* Re: Bug: git add :!x . exits with error when x is in .gitignore

https://lore.kernel.org/git/1d560aa1-d452-47f5-aaf2-4cb1ccdab100@gmail.com/
   Code review. Pointed out logical error.

* [PATCH v10] setup: allow cwd/.git to be a symlink to a directory

https://lore.kernel.org/git/20260220164512.216901-1-a3205153416@gmail.com/
   [Under review]
   After over half a month of discussions, repeated refactoring, and code
   reviews, I delved deep into setup.c. I gained insights into Git's 
design philosophy, and learned the art of striking a balance in 
developer communication. It took me a large amount of time and effort to 
thoroughly understand every line of the code. I often found myself 
poring over the call chain of a single function well into the night.... 
But I persevered until the end, and I believe my patience will see me 
through even larger projects.

ABOUT THE PROJECT -----------------

-- Synopsis

As far as I know, the Git community is actively working towards 'libification' - making Git's internal machinery reusable as a C library. The extensive reliance on global state is a major roadblock to this goal.

Many core functions implicitly read environment variables and store them in global static variables. This can cause several issues:

   1. Global variables prevent Git's core functions from being executed 
safely in multi-threaded contexts. For example, When unexpected states 
(e.g., a permission denied error when probing a directory), they often 
rely on the global state to decide whether to call die(), which 
internally calls exit(). It’s fine for a standalone CLI tool, but for a 
linked C library used by a long-running multi-threaded server, a single 
die() call will kill the entire host process. Structured status, instead 
of fatal exits, should be returned.
   2. When Git is called multiple times within the same process, global 
states can lead to memory leaks or incorrect behaviors.
   3. Unit testing becomes difficult because the environment must be 
artificially manipulated before calling functions.

Take a look at this example from environment.c:

     206 const char *get_commit_output_encoding(void)
     207 {
     208     return git_commit_encoding ? git_commit_encoding : "UTF-8";
     209 }

If Git is invoked as a C library by a multi-threaded server:
- Thread A formats a commit for Repo A (using GBK);
- Thread B concurrently formats a commit for Repo B (using UTF-8);

Then they will race to read and overwrite the exact same global `git_commit_encoding` pointer, which is not what we expect. Therefore, we have to refactor these environment variables by moving them from global scope into a well-defined and encapsulated context.

-- Approach

The task at hand goes beyond simply repackaging the global variables into the struct repository structure. Based on my recent experience refactoring setup.c, I realized that libification requires careful management of variable lifecycles and api boundaries:

     [ Current ]
     Core functions --------reads-------> Global variables (via getenv)
                                          [Thread unsafe]

     [ Target ]
     Core functions ----passes context--> struct repository
                                                 | owns
                                                 v
                                          struct repo_settings

				         other domain-specific structs

Although the principle is simple, the scope of changes is extensive. The following three-step approach can serve as a guiding principle for it:

   1. Identify isolated environment variables currently residing in the
      global scope. Conduct a case-by-case analysis to map each variable
      to its most appropriate existing home (e.g., struct repo_settings
      for configuration values, or specific localized structs within
      struct repository).
   2. Instead of blindly passing struct repository *repo down into every
      single low-level library function, bubbling the dependency up is
      the true goal. External callers of the functions must be carefully
      audited to prevent regressions.
   3. Safely remove the old global variables and macro definitions. Make
      full use of Git's existing GitLab/GitHub CI and utilize local
      Meson builds with AddressSanitizer enabled to ensure that the new
      lifecycle introduces zero memory leaks.
   4. Many globals like `editor_program` are parsed once and remain
      available globally. New data flow might need to be designed to
      maintain the lazy-loading efficiency.

Additionally, given the anticipated high volume of commits, we must ensure each patch is independent and atomic, preventing any user-untraceable or unexplainable bugs from occurring in the codebase at any state.

AVAILABILITY ------------ Fortunately, my summer vacation coincides with the GSoC work period. I will treat this project as my primary focus, dedicating a minimum of 35 hours per week. If needed, I can work a 9-to-5 schedule.

I will have a significant head start to draft RFC patches before the official coding period even begins. Having this buffer period allows me to go through the rigorous code review process within the Git community with greater ease.

TIMELINE & MILESTONES --------------------- Considering the differences between this project and other projects on the idea list, rather than hoarding massive changes, I will submit 3-to-5-patch series frequently to respect reviewers' time and maintain a steady velocity.

Below is the tentative schedule I have prepared for myself:

* Community Bonding (May 1 - May 25): Planning & RFC
   - May 1 - May 7: Wrap up university finals. Discuss and finalize the
     prioritized list of subsystems with my mentor.
   - May 8 - May 25: Categorize the targeted global variables and map out
     their intended destinations (e.g., repo_settings). Draft and submit
     the initial RFC patch series.

* Phase 1 (May 26 - July 10): Foundation
   - Weeks 1-2: Plumb the context pointer (`struct repository *repo`) 
through call chains for simple variables (e.g., boolean flags or integer 
configs).
   - Weeks 3-4: Audit and update external callers to use the new API.
   - Weeks 5-6: Submit the first major refactoring patch series. Address
     mailing list feedback and resolve merge conflicts. (Midterm Evaluation)

* Phase 2 (July 11 - August 18): Complex Migration & Cleanup
   - Weeks 7-8: Refactor higher-complexity variables (e.g., path-related 
globals).
   - Weeks 9-10: Compile the codebase with AddressSanitizer and run the 
full test suite to execute strict memory leak checks.
   - Weeks 11-12: Remove unused global macro definitions and static 
variables. Update internal documentation and write the final GSoC report.

(The above is for reference only. Personally, I always finish tasks faster than planned 😉)

~$ git checkout HEAD@{postGSoC} ------------------------------- This past month since joining the Git community has been the most enjoyable month of my programming journey. To quote a close friend of mine (who is applying for the Neovim GSoC project):

   "Only fools chase trends; open source is the game for the brave."

The words may be blunt, but the logic holds true. This statement surely resonates with me (and maybe many other GSoC contributors): our passion for code and open-source drives us forward.

Even if I didn't make the cut, so what? ~$ git reset --hard... Just kidding. The Git codebase is far too interesting to abandon now.

------------------------------------------------------------------------- Changes since V3:

  - The idea of introducing a new container is abandoned now. Therefore, 
In approach section, the diagram and corresponding 
descriptions/"guidelines" are modified;
  - Emphasize on Meson and GitLab/GitHub which can be used for necessary 
safety checks;
  - Refined timeline section (community bonding).

  ** I wrote "Changes since v3" as well in last patch (V3). Sorry for 
the typo :( **

Regards,

Yuchen

Hi Tian

On 26/02/2026 17:02, Tian Yuchen wrote:

There are four steps below

Note that as settings in struct repo_settings are lazily parsed, it is only suitable for settings that are already lazily parsed. That means it is not a suitable home for any settings that are parsed at startup by git_default_config().

Where a function only needs one piece of information from struct repository that sounds like a good strategy.

Although `editor_program` is parsed once, that happens in git_default_config() so it is not lazily loaded and making it lazily loaded would be a regression as if the config value is invalid we want to exit with an error early in the process, not just before we prompt the user to edit a file.

Thanks

Phillip

Hi Phillip,

Wow, your reply is very detailed. Appreciate.

Yup, a typo.

This makes sense to me. So for variables in like 'git_default_config()', their startup parsing nature must be preserved. Will update the proposal to explicitly distinguish between these different lifecycles.

It's much better to pass just that value down rather than passing the entire 'struct repository', right?

Oh, I thought it was lazy-loaded. I completely overlooked the user experience in terms of a delayed fatal config error also. Will double check the source code and rewrite this part.

I'm delighted to see more people reviewing my proposal. I've truly gained valuable insights into Git's design philosophy. My sincere gratitude to you.

Regards,

Yuchen

Hi all,

Here is the V5 patch.

Thanks Phillip Wood, for help and guidance.

Refactoring in order to reduce Git's global state =================================================

PERSONAL INFORMATION
--------------------
Name: Tian Yuchen
E-mail: a3205153416@gmail.com
Phone number: +65 98740318
Time-zone: UTC + 08:00
Github: https://github.com/malon7782

Education: NTU, Singapore
Year: Year 1 semester 2
Degree: Electrical and Electronic Engineering (EEE)

PRE GSOC -------- I have always held a deep passion for the open-source community. Although I wasn't a computer science major, I tinkered with open-source projects long before college. I have solid hands-on experience in C programming and system-level debugging.

I use Ubuntu 24.04 on a daily basis, so I am proficient in using the Linux command line and CLI tools.

I have contributed to the Git community by sending patches. Since my first commit (17/1/2026), I have maintained a nearly daily contribution. Here is the list of contributions I have made:

* [PATCH v1] t1005: modernize "! test -f" to "test_path_is_missing"

https://lore.kernel.org/git/20260117062515.319664-1-a3205153416@gmail.com/
   This patch is my microproject, the first contribution I made to the 
codebase.
   [Graduated to 'master']

* [PATCH v2] t2203: avoid masking exit codes in git status

https://lore.kernel.org/git/20260118043537.338769-1-a3205153416@gmail.com/#t

* [PATCH v2] symlinks: use unsigned int for flags

https://lore.kernel.org/git/20260120152219.398999-1-a3205153416@gmail.com/
   [Will merge to 'next']

* [PATCH v4] t/perf/p3400: speed up setup using fast-import

https://lore.kernel.org/git/20260130170123.642344-1-a3205153416@gmail.com/
   [Will merge to 'master']

* Re: [PATCH] [RFC] attr: use local repository state in read_attr

https://lore.kernel.org/git/cc2f400e-49c2-4de0-9c51-9a5c0294735e@gmail.com/
   Code review. To verify the performance loss, I wrote a test script to
   measure the time difference before and after the modification.

* Re: Bug: git add :!x . exits with error when x is in .gitignore

https://lore.kernel.org/git/1d560aa1-d452-47f5-aaf2-4cb1ccdab100@gmail.com/
   Code review. Pointed out logical error.

* [PATCH v10] setup: allow cwd/.git to be a symlink to a directory

https://lore.kernel.org/git/20260220164512.216901-1-a3205153416@gmail.com/
   [Under review]
   After over half a month of discussions, repeated refactoring, and code
   reviews, I delved deep into setup.c. I gained insights into Git's 
design philosophy, and learned the art of striking a balance in 
developer communication. It took me a large amount of time and effort to 
thoroughly understand every line of the code. I often found myself 
poring over the call chain of a single function well into the night.... 
But I persevered until the end, and I believe my patience will see me 
through even larger projects.

ABOUT THE PROJECT -----------------

-- Synopsis

As far as I know, the Git community is actively working towards 'libification' - making Git's internal machinery reusable as a C library. The extensive reliance on global state is a major roadblock to this goal.

Many core functions implicitly read environment variables and store them in global static variables. This can cause several issues:

   1. Global variables prevent Git's core functions from being executed 
safely in multi-threaded contexts. For example, When unexpected states 
(e.g., a permission denied error when probing a directory), they often 
rely on the global state to decide whether to call die(), which 
internally calls exit(). It’s fine for a standalone CLI tool, but for a 
linked C library used by a long-running multi-threaded server, a single 
die() call will kill the entire host process. Structured status, instead 
of fatal exits, should be returned.

   2. When Git is called multiple times within the same process, global 
states can lead to memory leaks or incorrect behaviors.

   3. Unit testing becomes difficult because the environment must be 
artificially manipulated before calling functions.

Take a look at this example from environment.c:

     206 const char *get_commit_output_encoding(void)
     207 {
     208     return git_commit_encoding ? git_commit_encoding : "UTF-8";
     209 }

If Git is invoked as a C library by a multi-threaded server:
- Thread A formats a commit for Repo A (using GBK);
- Thread B concurrently formats a commit for Repo B (using UTF-8);

Then they will race to read and overwrite the exact same global `git_commit_encoding` pointer, which is not what we expect. Therefore, we have to refactor these environment variables by moving them from global scope into a well-defined and encapsulated context.

-- Approach

The task at hand goes beyond simply repackaging the global variables into the struct repository structure. Based on my recent experience refactoring setup.c, I realized that libification requires careful management of variable lifecycles and api boundaries:

     [ Current ]
     Core functions --------reads-------> Global variables (via getenv)
                                          [Thread unsafe]

     [ Target ]
     Core functions ----passes context--> struct repository
                                                 | owns
                                                 v
                                          struct repo_settings

                    	                 other domain-specific structs

Although the principle is simple, the scope of changes is extensive. The following insights can serve as a guiding principle for it:

   1. Identify isolated environment variables currently residing in the
      global scope. Conduct a case-by-case analysis to map each variable
      to its most appropriate existing home based on their lifecycles:

	Variables that are only parsed when needed will be safely mapped
	to struct repo_settings.

	Variables parsed at startup (e.g., editor_program)
	must not be moved to lazily parsed structs to ensure that
	invalid configurations can trigger early failures before
	execution proceeds too far, which is also for the sake of user
         experience.

   2. Instead of blindly passing struct repository *repo down into every
      single low-level library function, bubbling the dependency up is
      the true goal. External callers of the functions must be carefully
      audited to prevent regressions.

   3. Safely remove the old global variables and macro definitions. Make
      full use of Git's existing GitLab/GitHub CI and utilize local
      Meson builds with AddressSanitizer enabled to ensure that the new
      lifecycle introduces zero memory leaks.

Additionally, given the anticipated high volume of commits, we must ensure each patch is independent and atomic, preventing any user-untraceable or unexplainable bugs from occurring in the codebase at any state.

AVAILABILITY ------------ Fortunately, my summer vacation coincides with the GSoC work period. I will treat this project as my primary focus, dedicating a minimum of 35 hours per week. If needed, I can work a 9-to-5 schedule.

I will have a significant head start to draft RFC patches before the official coding period even begins. Having this buffer period allows me to go through the rigorous code review process within the Git community with greater ease.

TIMELINE & MILESTONES --------------------- Considering the differences between this project and other projects on the idea list, rather than hoarding massive changes, I will submit 3-to-5-patch series frequently to respect reviewers' time and maintain a steady velocity.

Below is the tentative schedule I have prepared for myself:

* Community Bonding (May 1 - May 25): Planning & RFC
   - May 1 - May 7: Wrap up university finals. Discuss and finalize the
     prioritized list of subsystems with my mentor.
   - May 8 - May 25: Categorize the targeted global variables and map out
     their intended destinations (e.g., repo_settings). Draft and submit
     the initial RFC patch series.

* Phase 1 (May 26 - July 10): Foundation
   - Weeks 1-2: Plumb the context pointer ('struct repository *repo') 
through call chains for simple variables (e.g., boolean flags or integer 
configs).
   - Weeks 3-4: Audit and update external callers to use the new API.
   - Weeks 5-6: Submit the first major refactoring patch series. Address
     mailing list feedback and resolve merge conflicts. (Midterm Evaluation)

* Phase 2 (July 11 - August 18): Complex Migration & Cleanup
   - Weeks 7-8: Refactor higher-complexity variables (e.g., path-related 
globals).
   - Weeks 9-10: Compile the codebase with AddressSanitizer and run the 
full test suite to execute strict memory leak checks.
   - Weeks 11-12: Remove unused global macro definitions and static 
variables. Update internal documentation and write the final GSoC report.

(The above is for reference only. Personally, I always finish tasks faster than planned 😉)

~$ git checkout HEAD@{postGSoC} ------------------------------- This past month since joining the Git community has been the most enjoyable month of my programming journey. To quote a close friend of mine (who is applying for the Neovim GSoC project):

   "Only fools chase trends; open source is the game for the brave."

The words may be blunt, but the logic holds true. This statement surely resonates with me (and maybe many other GSoC contributors): our passion for code and open-source drives us forward.

Even if I didn't make the cut, so what? ~$ git reset --hard... Just kidding. The Git codebase is far too interesting to abandon now.

------------------------------------------------------------------------- Changes since V4:

  - “Treating variables or functions differently based on their 
lifecycle” has been added to the Approach section.

  - Fixed a typo below the diagram.

Regards,

Yuchen

Hi Tian

On 27/02/2026 16:58, Tian Yuchen wrote:

The encoding config is really a user preference that lets the user compose commit messages in their preferred encoding while allowing git to store the message encoded as UTF-8. I'm struggling to see why two threads would be using different encodings as it implies that the user is using different encodings in different repositories.

Below you say

 >      Variables parsed at startup (e.g., editor_program)
 >      must not be moved to lazily parsed structs to ensure that
 >      invalid configurations can trigger early failures before
 >      execution proceeds too far, which is also for the sake of user
 >          experience.

i18n.commitEncoding is another such setting as it is currently eagerly parsed so I'm surprised to see it being converted to lazy parsing in https://lore.kernel.org/20260228190201.3684705-1-a3205153416@gmail.com

I'm afraid that the suggestion on the project webpage is not very helpful. Most config variables are unsuited to a conversion based on repository_settings, it would be better to look at the approach implemented in https://lore.kernel.org/48821a3848bef25c13038be8377ad73e7c17a924.1771258573.git.belkid98@gmail.com that is discussed in https://lore.kernel.org/xmqqwm1vk83a.fsf@gitster.g

Thanks

Phillip

Hi Phillip,

Thank you so much for taking the time to review both my patch and my GSoC proposal draft.

You are right on all points:

1. I now agree that commit encoding being a user/environment preference 
rather than a repository-specific attribution. My example was indeed a 
flawed assumption.

2. Thank you for catching the contradiction between my patch and my 
proposal's guiding principles. However, I changed it to lazy-loading not 
because I didn't follow the workflow I wrote, but because I thought the 
original eagerly parsing behavior was *incorrect*. But since I now 
understand point 1 above, this is no longer an issue to me anymore.

3. I deeply appreciate you pointing out that the GSoC ideas page might 
be misleading. I will study through the link you provide.

Thank you again for steering me in the right direction!

Regards,

Yuchen

Phillip Wood <phillip.wood123@gmail.com> writes:

Thanks for a dose of sanity here.  Very much appreciated.

Hi all,

Here is the V6 Draft. Looking forward to hearing your feedback (ゝ∀･)

Refactoring in order to reduce Git's global state =================================================

PERSONAL INFORMATION
--------------------
Name: Tian Yuchen
E-mail: a3205153416@gmail.com
Phone number: +65 98740318
Time-zone: UTC + 08:00
Github: https://github.com/malon7782

Education: NTU, Singapore
Year: Year 1 semester 2
Degree: Electrical and Electronic Engineering (EEE)

PRE GSOC -------- I have always held a deep passion for the open-source community. Although I'm not a computer science major, I tinkered with open-source projects long before college. I have solid hands-on experience in C programming and system-level debugging.

I use Ubuntu 24.04 on a daily basis, so I am proficient in using the Linux command line and CLI tools.

I have contributed to the Git community by sending patches. Since my first commit (17/1/2026), I have maintained a nearly daily contribution. Here is the list of contributions I have made:

* [PATCH v1] t1005: modernize "! test -f" to "test_path_is_missing"

https://lore.kernel.org/git/20260117062515.319664-1-a3205153416@gmail.com/
   This patch is my microproject, the first contribution I made to the 
codebase.
   [Graduated to 'master']

* [PATCH v2] t2203: avoid masking exit codes in git status

https://lore.kernel.org/git/20260118043537.338769-1-a3205153416@gmail.com/#t

* [PATCH v2] symlinks: use unsigned int for flags

https://lore.kernel.org/git/20260120152219.398999-1-a3205153416@gmail.com/
   [Merged to 'next']

* [PATCH v4] t/perf/p3400: speed up setup using fast-import

https://lore.kernel.org/git/20260130170123.642344-1-a3205153416@gmail.com/
   [Will merge to 'master']

* Re: [PATCH] [RFC] attr: use local repository state in read_attr

https://lore.kernel.org/git/cc2f400e-49c2-4de0-9c51-9a5c0294735e@gmail.com/
   Code review. To verify the performance loss, I wrote a test script to
   measure the time difference before and after the modification.

* Re: Bug: git add :!x . exits with error when x is in .gitignore

https://lore.kernel.org/git/1d560aa1-d452-47f5-aaf2-4cb1ccdab100@gmail.com/
   Code review. Pointed out logical error.

* [PATCH v11] setup: allow cwd/.git to be a symlink to a directory

https://lore.kernel.org/git/20260220164512.216901-1-a3205153416@gmail.com/
   [Under review]
   After over half a month of discussions, repeated refactoring, and code
   reviews, I delved deep into setup.c. I gained insights into Git's 
design philosophy, and learned the art of striking a balance in 
developer communication. It took me a large amount of time and effort to 
thoroughly understand every line of the code. I often found myself 
poring over the call chain of a single function well into the night.... 
But I persevered until the end, and I believe my patience will see me 
through even larger projects.

* [PATCH v4 0/3] move encoding configs to repo_config_values()

https://lore.kernel.org/git/20260228190201.3684705-1-a3205153416@gmail.com/
   [In progress]
   A practice patch for working according to the workflow described in 
this proposal.

* Re: [PATCH 4/4] repo: add the field path.toplevel

https://lore.kernel.org/git/e6e7e272-4aec-461e-aebd-33ec0a324770@gmail.com/
   Code review. Question unreasonable designs.

ABOUT THE PROJECT -----------------

-- Synopsis

As far as I know, the Git community is actively working towards 'libification' - making Git's internal machinery reusable as a C library. The extensive reliance on global state is a major roadblock to this goal.

Many core functions implicitly read environment variables and store them in global static variables. This can cause several issues:

   1. When Git is called multiple times within the same process, global 
states can lead to memory leaks or incorrect behaviors.

   2. Unit testing becomes difficult because the environment must be 
artificially manipulated before calling functions.

   3. Global variables prevent Git's core functions from being executed 
safely in multi-threaded contexts. For example, When unexpected states 
(e.g., a permission denied error when probing a directory), they often 
rely on the global state to decide whether to call die(), which 
internally calls exit(). It’s fine for a standalone CLI tool, but for a 
linked C library used by a long-running multi-threaded server, a single 
die() call will kill the entire host process. Structured status, instead 
of fatal exits, should be returned.

Take a look at this example from environment.c:

     206 const char *get_commit_output_encoding(void)
     207 {
     208     return git_commit_encoding ? git_commit_encoding : "UTF-8";
     209 }

If Git is invoked as a C library by a multi-threaded server:
- Thread A formats a commit for Repo A (using GBK);
- Thread B concurrently formats a commit for Repo B (using UTF-8);

Then they will race to read and overwrite the exact same global `git_commit_encoding` pointer, which is not what we expect. Therefore, we have to refactor these environment variables by moving them from global scope into a well-defined and encapsulated context.

-- Approach

The task at hand goes beyond simply repackaging the global variables into the struct repository structure. Based on my recent experience refactoring setup.c, I realized that libification requires careful management of variable lifecycles and api boundaries:

     [ Current ]
     Core functions --------reads-------> Global variables (via getenv)
                                          [Thread unsafe]

     [ Target ]
     Core functions ----passes context--> struct repository
                                                 | owns
                                                 v
                                      struct repo_settings(lazy)

		  	          struct repo_config_values (eager) [1]

                                      other domain-specific structs

Although the principle is simple, the scope of changes is extensive. The following insights can serve as a guiding principle for it:

   1. Identify isolated environment variables currently residing in the
      global scope. Conduct a case-by-case analysis to map each variable
      to its most appropriate existing home based on their lifecycles:

     	Variables that are only parsed when needed will be safely mapped
	to struct repo_settings.

     	Variables parsed at startup (e.g., editor_program) must not be
	moved to lazily parsed structs to ensure that invalid
	configurations can trigger early failures before execution
	proceeds too far, which is also for the sake of user experience.
	(Phillip Wood points out that the struct repo_config_values()
	can serve as a good home to these variables, though this
	approach remains in its early stages and has not yet been fully
	confirmed and implemented. [2])

   2. Instead of blindly passing struct repository *repo down into every
      single low-level library function, bubbling the dependency up is
      the true goal. External callers of the functions must be carefully
      audited to prevent regressions.

   3. Safely remove the old global variables and macro definitions. Make
      full use of Git's existing GitLab/GitHub CI and utilize local
      Meson builds with AddressSanitizer enabled to ensure that the new
      lifecycle introduces zero memory leaks. [3]

Additionally, given the anticipated high volume of commits, we must ensure each patch is independent and atomic [4], preventing any user-untraceable or unexplainable bugs from occurring in the codebase at any state.

AVAILABILITY ------------ Fortunately, my summer vacation perfectly coincides with the GSoC work period. I will treat this project as my primary focus, dedicating a minimum of 35 hours per week. If needed, I can work a 9-to-5 schedule.

I will have a significant head start to draft RFC patches before the official coding period even begins. Having this buffer period allows me to go through the rigorous code review process within the Git community with greater ease.

TIMELINE & MILESTONES --------------------- Considering the differences between this project and other projects on the idea list, rather than hoarding massive changes, I will submit 3-to-5-patch series frequently to respect reviewers' time and maintain a steady velocity.

Below is the tentative schedule I have prepared for myself:

* Community Bonding (May 1 - May 25): Planning & RFC
   - May 1 - May 7: Wrap up university finals. Discuss and finalize the
     prioritized list of subsystems with my mentor.
   - May 8 - May 25: Categorize the targeted global variables and map out
     their intended destinations (e.g., repo_settings vs 
repo_config_values). Draft and submit
     the initial RFC patch series.

* Phase 1 (May 26 - July 10): Foundation
   - Weeks 1-2: Plumb the context pointer ('struct repository *repo') 
through call chains for simple variables (e.g., boolean flags or integer 
configs).
   - Weeks 3-4: Audit and update external callers to use the new API.
   - Weeks 5-6: Submit the first major refactoring patch series. Address
     mailing list feedback and resolve merge conflicts. (Midterm Evaluation)

* Phase 2 (July 11 - August 18): Complex Migration & Cleanup
   - Weeks 7-8: Refactor higher-complexity variables (e.g., path-related 
globals).
   - Weeks 9-10: Compile the codebase with AddressSanitizer and run the 
full test suite to execute strict memory leak checks.
   - Weeks 11-12: Remove unused global macro definitions and static 
variables. Update internal documentation and write the final GSoC report.

(The above is for reference only. Personally, I always finish tasks faster than planned 😉)

~$ git checkout HEAD@{postGSoC} ------------------------------- This past month since joining the Git community has been the most enjoyable month of my programming journey. To quote a close friend of mine (who is applying for the Neovim GSoC project):

   "Only fools chase trends; open source is the game for the brave."

The words may be blunt, but the logic holds true. This statement surely resonates with me (and maybe many other GSoC contributors): our passion for code and open-source drives us forward.

Even if I didn't make the cut, so what? ~$ git reset --hard... Just kidding. The Git codebase is far too interesting to abandon now.

REFERENCE ------------------------------- [1]

https://lore.kernel.org/all/48821a3848bef25c13038be8377ad73e7c17a924.1771258573.git.belkid98@gmail.com/

[2]

https://lore.kernel.org/git/CAP8UFD2Q7gctwzGOe+rbgdXZSbDbV0dmM-cx4qt_d8nKi88=HA@mail.gmail.com/T/#t

[3]

https://lore.kernel.org/all/CAOLa=ZR=2B7yH+vtyiAPcCyU17yd2GZwonaj=JRo1f+LzSCoTg@mail.gmail.com/

[4]

https://lore.kernel.org/all/xmqqy0kp7wai.fsf@gitster.g/

------------------------------------------------------------------------- Changes since V5:

  - Once again, the diagram and approach sections emphasize 
distinguishing variables across different life cycles.

  - Modified the --Synopsis section. (issues with global variables)

  - Included recent contributions and updated progress.

  - Reference links has been added at the end to clarify the source of 
my viewpoint/plan.

Regards,

Yuchen

Hi all,

Here is the V7 Draft. Looking forward to hearing your feedback (ゝ∀･)

Refactoring in order to reduce Git's global state =================================================

PERSONAL INFORMATION
--------------------
Name: Tian Yuchen
E-mail: a3205153416@gmail.com
Phone number: +65 98740318
Time-zone: UTC + 08:00
Github: https://github.com/malon7782

Education: NTU, Singapore
Year: Year 1 semester 2
Degree: Electrical and Electronic Engineering (EEE)

PRE GSOC -------- I have always held a deep passion for the open-source community. Although I'm not a computer science major, I tinkered with open-source projects long before college. I have solid hands-on experience in C programming and system-level debugging.

I use Ubuntu 24.04 on a daily basis, so I am proficient in using the Linux command line and CLI tools.

I have contributed to the Git community by sending patches. Since my first commit (17/1/2026), I have maintained a nearly daily contribution. Here is the list of contributions I have made:

* [PATCH v1] t1005: modernize "! test -f" to "test_path_is_missing"

https://lore.kernel.org/git/20260117062515.319664-1-a3205153416@gmail.com/
   This patch is my microproject, the first contribution I made to the 
codebase.
   [Graduated to 'master']

* [PATCH v2] t2203: avoid masking exit codes in git status

https://lore.kernel.org/git/20260118043537.338769-1-a3205153416@gmail.com/#t

* [PATCH v2] symlinks: use unsigned int for flags

https://lore.kernel.org/git/20260120152219.398999-1-a3205153416@gmail.com/
   [Merged to 'next']

* [PATCH v4] t/perf/p3400: speed up setup using fast-import

https://lore.kernel.org/git/20260130170123.642344-1-a3205153416@gmail.com/
   [Will merge to 'master']

* Re: [PATCH] [RFC] attr: use local repository state in read_attr

https://lore.kernel.org/git/cc2f400e-49c2-4de0-9c51-9a5c0294735e@gmail.com/
   Code review. To verify the performance loss, I wrote a test script to
   measure the time difference before and after the modification.

* Re: Bug: git add :!x . exits with error when x is in .gitignore

https://lore.kernel.org/git/1d560aa1-d452-47f5-aaf2-4cb1ccdab100@gmail.com/
   Code review. Pointed out logical error.

* [PATCH v11] setup: allow cwd/.git to be a symlink to a directory

https://lore.kernel.org/git/20260220164512.216901-1-a3205153416@gmail.com/
   [Under review]
   After over half a month of discussions, repeated refactoring, and code
   reviews, I delved deep into setup.c. I gained insights into Git's 
design philosophy, and learned the art of striking a balance in 
developer communication. It took me a large amount of time and effort to 
thoroughly understand every line of the code. Throughout this process, I 
meticulously examined portions of the call chain in setup.c, the timing 
of die() usage, expected lookup/error handling behavior, external 
callers, and ran several full CI tests. I managed to correct my initial 
assumptions and it taught me why die() cannot be casually bypassed in 
libification without rigorous error propagation, as well as how to 
handle cross-platform CI edge cases.

* [PATCH v4 0/3] move encoding configs to repo_config_values()

https://lore.kernel.org/git/20260228190201.3684705-1-a3205153416@gmail.com/
   [In progress]
   A practice patch for working according to the workflow described in 
this proposal.

* Re: [PATCH 4/4] repo: add the field path.toplevel

https://lore.kernel.org/git/e6e7e272-4aec-461e-aebd-33ec0a324770@gmail.com/
   Code review. Questioned unreasonable designs.

* Re: [PATCH] Refactor 'trust_executable_bit' to repository-scoped setting

https://lore.kernel.org/git/24f40e5a-a5fd-49ec-86e7-921b44e4abd9@gmail.com/
   Code review. Clarify lazy-loading and eager-loading. Pointed out 
problems with the patch.

* [PATCH] patch-ids: achieve const correctness in patch_id_neq()
   https://lore.kernel.org/git/xmqqseaasuph.fsf@gitster.g/T/#t
   [In progress]
   Discuss and modify the items marked NEEDSWORK in patch-ids.c

To date, developing diverse patches across various domains has been a thoroughly enjoyable experience for me. I relish exploring different fields while maintaining the discipline to delve deeply into them. I never settle for superficial patches.

ABOUT THE PROJECT -----------------

-- Synopsis

As far as I know, the Git community is actively working towards 'libification' (laid by Patrick Steinhardt's config/path refactoring, and Olamide Bello's recent introduction of repo_config_values) - making Git's internal machinery reusable as a C library. The extensive reliance on global state is a major roadblock to this goal.

Many core functions implicitly read environment variables and store them in global static variables. This can cause several issues:

   1. When Git is called multiple times within the same process, global 
states can lead to memory leaks or incorrect behaviors.

   2. Unit testing becomes difficult because the environment must be 
artificially manipulated before calling functions.

   3. Global variables prevent Git's core functions from being executed 
safely in multi-threaded contexts. For example, when encountering 
unexpected states (e.g., a permission denied error when probing a 
directory), core functions often rely on the global state to decide 
whether to call die(), which internally calls exit(). It’s fine for a 
standalone CLI tool, but for a linked C library used by a long-running 
multi-threaded server, a single die() call will kill the entire host 
process. Structured status, instead of fatal exits, should be returned.

Take a look at this example from environment.c:

     206 const char *get_commit_output_encoding(void)
     207 {
     208     return git_commit_encoding ? git_commit_encoding : "UTF-8";
     209 }

If Git is invoked as a C library by a multi-threaded server:
- Thread A formats a commit for Repo A (using GBK);
- Thread B concurrently formats a commit for Repo B (using UTF-8);

Then they will race to read and overwrite the exact same global `git_commit_encoding` pointer, which is not what we expect. Therefore, we have to refactor these environment variables by moving them from global scope into a well-defined and encapsulated context.

-- Trade-off?

A naive approach to achieve this is blindly moving every global variable into struct repository. For the previous encoding case, this approach does not seem entirely unfeasible. However, it is easy to cite another example:

“What if some variables are accessed before setup_git_directory() successfully initializes the repo struct?”

For instance, early discovery flags like is_bare_repository_cfg are assigned during the .git directory probe, long before a struct repository object can be safely instantiated. We create a chicken-and-egg initialization paradox if we force these early-startup variables into repo_settings. How can a variable dictate the discovery of a repository if it lives inside the repository it is trying to discover?

I believe the key point to note for this case is the boundary between Repository State and Process/Startup State. Encapsulating it within the early-boot context might be a viable option... but more importantly, this example illustrates that eliminating global variables is never a matter of blindly “moving … to …” Instead, it requires careful, multifaceted consideration and a thorough weighing of the trade-offs.

-- Approach

Therefore, the task at hand goes beyond simply repackaging the global variables into the struct repository structure. Based on my recent experience refactoring setup.c, I realized that libification requires careful management of variable lifecycles and api boundaries:

     [ Current ]
     Core functions --------reads-------> Global variables (via getenv)
                                          [Thread unsafe]

     [ Target ]
     Core functions ----passes context--> struct repository
                                                 | owns
                                                 v
                                      struct repo_settings(lazy)

                                 struct repo_config_values (eager) [1]

                                      other domain-specific structs

Although the principle is simple, the scope of changes is extensive. The following insights can serve a guiding principles (but not the absolute rules to obey):

   1. Identify isolated environment variables currently residing in the
      global scope. Conduct a case-by-case analysis to map each variable
      to its most appropriate existing home based on their lifecycles:

         Variables that are only parsed when needed will be safely mapped
     to struct repo_settings.

         Variables parsed at startup (e.g., editor_program) must not be
     moved to lazily parsed structs to ensure that invalid
     configurations can trigger early failures before execution
     proceeds too far, which is also for the sake of user experience.
     (Phillip Wood points out that the struct repo_config_values
     can serve as a good home to these variables, though this
     approach remains in its early stages and has not yet been fully
     confirmed and implemented. [2])

   2. Instead of blindly passing struct repository *repo down into every
      single low-level library function, bubbling the dependency up is
      the true goal. External callers of the functions must be carefully
      audited to prevent regressions.

   3. Safely remove the old global variables and macro definitions. Make
      full use of Git's existing GitLab/GitHub CI and utilize local
      Meson builds with AddressSanitizer enabled to ensure that the new
      lifecycle introduces zero memory leaks. [3]

Additionally, given the anticipated high volume of commits, we must ensure each patch is independent and atomic [4], preventing any user-untraceable or unexplainable bugs from occurring in the codebase at any state.

AVAILABILITY ------------ Fortunately, my summer vacation perfectly coincides with the GSoC work period. I will treat this project as my primary focus, dedicating a minimum of 35 hours per week. If needed, I can work a 9-to-5 schedule.

I will have a significant head start to draft RFC patches before the official coding period even begins. Having this buffer period allows me to go through the rigorous code review process within the Git community with greater ease.

I've always kept up the habit of blogging (though previously it was mostly literary essays and musings). For this GSoC project, I'll provide regular progress updates (every week or every other week). I'll also occasionally share technical insights.

TIMELINE & MILESTONES --------------------- I believe that outlining a rigid, day-by-day schedule months in advance is unrealistic for a breathing codebase like Git. I will employ a pipeline-driven workflow:

   1. Small refactoring every 3-5 days, major refactoring every 2-3 
weeks, alternating between the two.
   2. After gathering sufficient, well-directed review suggestions, 
revisit and modify the previous refactoring.
   3. Ensure these patches do not depend on one another to prevent a 
domino effect.

Below is the tentative schedule I have prepared for myself:

* Community Bonding (May 1 - May 25): Planning & RFC
   - May 1 - May 7: Wrap up university finals. Discuss and finalize the
     prioritized list of subsystems with my mentor.
   - May 8 - May 25: Categorize the targeted global variables and map out
     their intended destinations (e.g., repo_settings vs 
repo_config_values). Draft and submit
     the initial RFC patch series.

* Phase 1 (May 26 - July 10): Foundation

   - Weeks 1-3: Target straightforward boolean flags and integer configs 
in environment.c (e.g., is_bare_repository_cfg). Plumb the context 
pointer, adapt callers, and dispatch the first 2-3 patch series.
   - Weeks 4-6: Process mailing list feedback and iterate (v2/v3). 
Concurrently begin migrating the next batch of variables. By the midterm 
evaluation, a steady rhythm of proposing, reviewing, and merging should 
be established.

* Phase 2 (July 11 - August 18): Complex Migration & Cleanup

   - Weeks 7-8: Shift focus to higher-complexity variables and string 
configurations (e.g., editor_program, comment_line_str). Apply the 
repo_config_values API where eager loading is strictly required.
   - Weeks 9-10: Finalize the remaining targeted globals. Conduct a 
global audit for any dangling macro definitions (like 
USE_THE_REPOSITORY_VARIABLE in fully refactored subsystems) and 
eliminate them.
   - Weeks 11-12: Update internal documentation 
(Documentation/technical/), ensure all patch series are either merged or 
in a stable state on the next branch, and write the final GSoC report.

~$ git checkout HEAD@{postGSoC} ------------------------------- I plan to stay active in the community long after the summer ends — reviewing related patches, mentoring future newcomers, and seeing the libification effort through to the end.

This past month since joining the Git community has been the most enjoyable month of my programming journey. To quote a close friend of mine (who is applying for the Neovim GSoC project):

   "Only fools chase trends; open source is the game for the brave."

The words may be blunt, but the logic holds true. This statement surely resonates with me (and maybe many other GSoC contributors): our passion for code and open-source drives us forward.

Even if I didn't make the cut, so what? ~$ git reset --hard…

Jokes aside, diving into Git's codebase this past month has been immensely rewarding. Win or lose, I'm here to stay and help.

REFERENCE ------------------------------- [1]

https://lore.kernel.org/all/48821a3848bef25c13038be8377ad73e7c17a924.1771258573.git.belkid98@gmail.com/

[2]

https://lore.kernel.org/git/CAP8UFD2Q7gctwzGOe+rbgdXZSbDbV0dmM-cx4qt_d8nKi88=HA@mail.gmail.com/T/#t

[3]

https://lore.kernel.org/all/CAOLa=ZR=2B7yH+vtyiAPcCyU17yd2GZwonaj=JRo1f+LzSCoTg@mail.gmail.com/

[4]

https://lore.kernel.org/all/xmqqy0kp7wai.fsf@gitster.g/

------------------------------------------------------------------------- Changes since V6:

  - Update recent contributions. Rewrite descriptions for some of the 
contributions.

  - Add a section on trade-offs, discussing insights gained from 
challenging scenarios I may encounter.

  - Completely rewrite the schedule.

  - Some expressions have been modified.

Regards,

Yuchen

I really hope someone can offer some advice! After all, the deadline is fast approaching...

By the way, I was wondering how the mentors review proposals. Can we revise and resubmit them multiple times? Is it better to submit the proposal as early as possible?

I’ve made some final (and quite significant) adjustments. Compared to v7, it incorporates more personal insights and methodologies. Here’s a link to the Google Doc.

https://docs.google.com/document/d/1t2sznOvnPz-9tOzVMH--pLxzRqYSJCFzqVWBVfL_NP8/edit?tab=t.0#heading=h.c3c40ftj1ilv

Regards,

Yuchen