Branching

[philocalyst]

This Pull Request fixes/closes #81.

It changes the following:

• Implements a complete view of commit lineage
• Borrows the logic/implementation from the well-tested guitar
• Conforms to the streaming model that was used in the previous version of the displays, meaning its performance impact is negligible.
• Has a new keybind to hide the graph, as a toggle.
• Expands and contracts dynamically based on what is currently on-screen
• Makes good use of space by compressing whitespace when possible (Through re-using lanes)
• Remains consistent against the rest of the UI.

I followed the checklist:

• I added unittests
• I ran make check without errors
• I tested the overall application
• I added an appropriate item to the changelog

[philocalyst]

@extrawurst Would love your feedback on this :) Quite large.

OH! And I just noticed that I removed the missing docs lint, I'll add that back if we agree on this general direction.

[gogongxt]

LGTM

[extrawurst]

please match the ususal format

[extrawurst]

I think we need to improve the visualization still, this is the same point in time of the history in three different tools: gitui, tig and fork:

[philocalyst]

Being precise here, the only difference I can see in the lack of a bottom connector? Or is there something else? ▰▰▰▰▰ Miles Wirth 🙃

…

From: extrawurst ***@***.***> Sent: 29 March 2026 11:53 To: gitui-org/gitui ***@***.***> Cc: Miles Wirht ***@***.***>, Author ***@***.***> Subject: Re: [gitui-org/gitui] Branching (PR #2890) extrawurst left a comment (gitui-org/gitui#2890) I think we need to improve the visualization still, this is the same point in time of the history in three different tools: gitui, tig and fork: <img width="623" height="291" alt="Screenshot 2026-03-29 at 19 50 36" src="https://github.com/user-attachments/assets/a05fed09-ce18-4f55-8915-e143b37b01d3" /> <img width="392" height="326" alt="Screenshot 2026-03-29 at 19 50 27" src="https://github.com/user-attachments/assets/8669dbeb-604b-4d25-ba8c-de5aa701ac95" /> <img width="895" height="227" alt="Screenshot 2026-03-29 at 19 51 51" src="https://github.com/user-attachments/assets/5cf8535b-3ce6-44a2-944d-a33c81c548ce" /> -- Reply to this email directly or view it on GitHub: #2890?email_source=notifications&email_token=A3MQBNB5QFY6DKIO5JS2ZSD4TFWKZA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTIMJVGA4DKNBWGE32M4TFMFZW63VGMF2XI2DPOKSWK5TFNZ2LK4DSL5RW63LNMVXHIX3POBSW4X3DNRUWG2Y#issuecomment-4150854617 You are receiving this because you authored the thread. Message ID: ***@***.***>

[VojtaStanek]

I am looking forward to this change!

By a quick look the source it seems like it works only with merge commits with 2 parents, but commit can have an arbitrary number of parents (see e.g. here).

[philocalyst]

@VojtaStanek Yes! Octopus merges! That's not covered here.

I couldn't find an implementation that handles them in a way that maps well to streaming. They don't crash anything, and happen incredibly rarely. If someone is able to extend what I have here, they're welcome to; but it shouldn't block merging.

[cruessler]

Thanks a lot for getting this started, that’s much appreciated! I just started reviewing the changes, but then I had a few more general questions.

In its current form, I find it hard to see merge commits and how they connect to their respective parents. The image below shows a comparison between this branch and tig in the gitoxide repository that has quite a few merge commits. Is there anything I need to set up to make merge commits more prominent? Does my terminal font need to have certain glyphs?

Also, in tig, vertical lines don’t get cut by horizontal ones, as opposed to guitar. I think I like tig’s approach better. What do you think?

[philocalyst]

No, that's just how they look :sad:. Good points -- I also wonder if people like the pressed texts to the facets of the branch lines? I could implement that as well. I think symbols... look cooler; but there are readability improvements we could aim for regardless, that these screenshots are making clear to me. The glyphs could expose be exposed as configuration. ▰▰▰▰▰ Miles Wirth 🙃

…

From: Christoph Rüßler ***@***.***> Sent: 07 April 2026 12:37 To: gitui-org/gitui ***@***.***> Cc: Miles Wirht ***@***.***>, Author ***@***.***> Subject: Re: [gitui-org/gitui] Branching (PR #2890) cruessler left a comment (gitui-org/gitui#2890) Thanks a lot for getting this started, that’s much appreciated! I just started reviewing the changes, but then I had a few more general questions. In its current form, I find it hard to see merge commits and how they connect to their respective parents. The image below shows a comparison between this branch and `tig` in the `gitoxide` repository that has quite a few merge commits. Is there anything I need to set up to make merge commits more prominent? Does my terminal font need to have certain glyphs? Also, in `tig`, vertical lines don’t get cut by horizontal ones, as opposed to `guitar`. I think I like `tig`’s approach better. What do you think? <img width="3440" height="1412" alt="Screenshot_20260407_212543" src="https://github.com/user-attachments/assets/c224b4a6-6a01-4942-be2f-fef91d63d40a" /> -- Reply to this email directly or view it on GitHub: #2890 (comment) You are receiving this because you authored the thread. Message ID: ***@***.***>

[extrawurst]

I also wonder if people like the pressed texts to the facets of the branch lines? I could implement that as well.

Not for now.

The glyphs could expose be exposed as configuration.

Also not relevant for now. Lets find a visual style thats a good default and add customization in a later PR. THis one is big enough as is.

In its current form, I find it hard to see merge commits and how they connect to their respective parents

I agree with this. we either find better symbols or go for the letter M.

Furthermore I think we should use different colored lines for parallel branches like tig does for more clear separation.

[philocalyst]

With the updates, the PR is now looking like the above. Is this a better direction? @extrawurst

I plan to submit a follow-up PR with selection-based highlighting which I think will clear up the remaining vie
wing complexities (As ascii characters can only show so much, realistically). This would look like cutting over the "background" lines when a particular lineage is focused.

[balintbarna]

Hi all, if you need a reference for what looks good and works well in TUI then lazygit (go) and git-igitt (rust) are both very good at visualizing the graph. I think the table layout of lazygit makes more sense:

short hash | timestamp | name | graph | (tags, branches) message

it can also switch between --all mode and current branch
and the colors and highlighting are great
I think it would be a good source of inspiration for the desired look
it also has multi-commit selection by holding down shift and a lot of useful shortcuts on this view, such as squash, fixup, drop, edit, amend, checkout, reset, cherry-pick, which is probably a separate issue from visualizing it, but mentioned it for completeness sake.

https://github.com/jesseduffield/lazygit#commit-graph
https://github.com/git-bahn/git-igitt

[philocalyst]

When I looked at the algo's of git-gitt there was some issues around their design as a second-pass over a complete input, which is counter to the streaming design of gitui. Definitely still some issues I realize I miss when I review, I think there's still some issues with connecting pipes, but the tradeoffs of the algorithm feel solid to me.

[balintbarna]

@philocalyst thanks for your reply. I don't have any insight on the code of the above mentioned projects, my comment is purely from a user perspective.

I'm a software professional and my main way of interacting with git is lazygit, before that I used the integration in vs code with the very popular git graph extension and before that I used source tree. I also briefly tried gitkraken and more recently gitui. I use the CLI when I need to or when it's simpler.

While lazygit has some serious bugs and I disagree with some of their decisions which make me miss git graph from vs code, it is my favourite tool so far, and a big part of that is their graph view, and even if the first version in gitui isn't equivalent, I would like to suggest it as inspiration for what it could become, in addition to the also fantastic git graph extension in vs code and the visually appealing git-igitt.

[philocalyst]

I'm with you !! Yeah I do plan to do follow-up PRs, just adding in more of these complex previews might have to be done on a second pass on just the active window. ▰▰▰▰▰ Miles Wirth 🙃

…

From: balintbarna ***@***.***> Sent: 22 April 2026 15:05 To: gitui-org/gitui ***@***.***> Cc: Miles Wirht ***@***.***>, Mention ***@***.***> Subject: Re: [gitui-org/gitui] Branching (PR #2890) balintbarna left a comment (gitui-org/gitui#2890) @philocalyst thanks for your reply. I don't have any insight on the code of the above mentioned projects, my comment is purely from a user perspective. I'm a software professional and my main way of interacting with git is lazygit, before that I used the integration in vs code with the very popular git graph extension and before that I used source tree. I also briefly tried gitkraken and more recently gitui. I use the CLI when I need to or when it's simpler. While lazygit has some serious bugs and I disagree with some of their decisions which make me miss git graph from vs code, it is my favourite tool so far, and a big part of that is their graph view, and even if the first version in gitui isn't equivalent, I would like to suggest it as inspiration for what it could become. -- Reply to this email directly or view it on GitHub: #2890?email_source=notifications&email_token=A3MQBNCKMUOFIA5ZN45NX3L4XE64LA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTIMZQGAZDKMRUHEYKM4TFMFZW63VHNVSW45DJN5XKKZLWMVXHJNLQOJPWG33NNVSW45C7N5YGK3S7MNWGSY3L#issuecomment-4300252490 You are receiving this because you were mentioned. Message ID: ***@***.***>

[cruessler]

@philocalyst Thanks for the changes! I’ve now started the proper first review which should be ready by today or tomorrow if all goes according to plan! 😀

[philocalyst]

Thanks so much :) ▰▰▰▰▰ Miles Wirth 🙃

…

From: Christoph Rüßler ***@***.***> Sent: 26 April 2026 23:24 To: gitui-org/gitui ***@***.***> Cc: Miles Wirht ***@***.***>, Mention ***@***.***> Subject: Re: [gitui-org/gitui] Branching (PR #2890) cruessler left a comment (gitui-org/gitui#2890) @philocalyst Thanks for the changes! I’ve now started the proper first review which should be ready by today or tomorrow if all goes according to plan! 😀 -- Reply to this email directly or view it on GitHub: #2890 (comment) You are receiving this because you were mentioned. Message ID: ***@***.***>

[peso]

Hi there. I'm still working on getting git-graph, now gleisbau, into gitui. The speed requirement from gitui can be solved by splitting the global branch trace algorithm off to a background thread, and then start at head. This will feel as if it starts instantly, but it will traverse the entire repo in due time. The split also addresses the memory issue seen in the gitui+gitgraph prototype from about a year go.

This is a major refactoring, so it will take some time. git-igitt is part of the git-bahn family and uses git-graph/gleisbau under the hood.

I think it is a good idea to get a graph feature into gitui as soon as possible. Everybody wants it. Maybe gleisbau can produce a visual improvement, but the major step is simply getting a graph. Congratulations to @philocalyst for reaching that goal.

[cruessler]

I’m sorry it’s taking so long! In order to not make you wait longer, I’m now sending you my first batch of comments, hoping they are already useful. As soon as I’m done with the next batch, you’ll get the next round of comments. I hope that works for you!

[philocalyst]

Balls back in your court! @cruessler

[cruessler]

@philocalyst wrote:
Balls back in your court! @cruessler

Really sorry it’s taking so long, I should be able to devote some time to this PR this week as I don’t have too many other obligations.

[philocalyst]

@cruessler Appreciated! Have another PR I've been sitting on for LFS, but didn't want to overwhelm the team -- so I'll be monitoring my github notifications closely to make sure this week isn't wasted!!

[cruessler]

I’m slowly getting myself familiar with the PR, in order to make sure I actually understand myself what is going on. 😀 I’ve got a couple of questions and remarks, some more fundamental than others. I’m also going to push minor refactorings that don’t warrant a comment to your branch as well, in order to make some areas more idiomatic. I hope that’s ok, if not, please let me know!

[philocalyst]

okay @cruessler you inspired me to do a pretty big refactor so sorry for the delay here! put a ton of effort into reducing the amount of redundant work as far as both clones and passing through the same commits/lanes twice or more. should be totally lazy as far as the graph goes, which i'm very happy to report! there's also a set of tests that should test the basics... (disclosure that i used $llm for the test writing, let me know if this goes against a policy and i will rewrite!)

Also I kind of overpromised with stashes, although I will do in a follow-up if all goes well :)

[andras-tim]

sorry for the offtopic, but I can't wait this feature for months... 😓
Could you please to merge it if there's no serious issues? This is an awesome feature that would make the gitui tui complete! (I'm going to finally switch from git-gui & gitk to gitui 😍)

[philocalyst]

Will fortune and fate it should be complete, just waiting on final review, btw @cruessler that force push is with lease, it was just a cargo.toml change! Please don't fret.

[cruessler]

This PR is on top of my TODO list for the coming week! I have drafted a benchmark to make sure there’s no performance regressions, but I have not yet gotten to committing and pushing that code.

[cruessler]

@philocalyst The benchmark is available in this commit: cruessler@d372bfa. This commit can be applied on top of your PR. The benchmark exercises a history walk with and without creating the history graph. Would you mind double-checking that it actually measures your changes? (I had it drafted by an LLM and want to make sure I didn’t miss anything in my review.)

Also, could you run the part without graph creation on your machine and compare against master? (This requires some modifications to the benchmark, but should not be too complicated.)

Example result on my machine:

❯ GITUI_BENCH_REPO=~/github/extrawurst/gitui/ cargo bench -p asyncgit --bench full_history
    Finished `bench` profile [optimized] target(s) in 0.13s
     Running benches/full_history.rs (target/release/deps/full_history-1e929556af4de21d)
full_history/without_graph
                        time:   [17.477 ms 17.660 ms 17.786 ms]
                        thrpt:  [146.18 Kelem/s 147.23 Kelem/s 148.77 Kelem/s]
                 change:
                        time:   [+0.4870% +1.6229% +2.7906%] (p = 0.02 < 0.05)
                        thrpt:  [−2.7148% −1.5970% −0.4847%]
                        Change within noise threshold.
full_history/with_graph time:   [35.523 ms 35.901 ms 36.392 ms]
                        thrpt:  [71.444 Kelem/s 72.421 Kelem/s 73.193 Kelem/s]
                 change:
                        time:   [−0.2869% +1.1908% +2.8456%] (p = 0.17 > 0.05)
                        thrpt:  [−2.7669% −1.1768% +0.2878%]
                        No change in performance detected.
Found 1 outliers among 10 measurements (10.00%)
  1 (10.00%) high mild

So it looks like creating the graph requires about the same amount of time as walking the history in this example. Running the full_history/without_graph part on master requires about the same amount of time as on this branch, so it looks like your changes don’t affect the “graph is disabled”-case in this synthetic benchmark.

❯ GITUI_BENCH_REPO=~/github/extrawurst/gitui/ cargo bench -p asyncgit --bench full_history
    Finished `bench` profile [optimized] target(s) in 0.10s
     Running benches/full_history.rs (target/release/deps/full_history-164177db5604f630)
full_history/without_graph
                        time:   [17.510 ms 17.691 ms 17.808 ms]
                        thrpt:  [146.01 Kelem/s 146.97 Kelem/s 148.49 Kelem/s]
                 change:
                        time:   [+0.3010% +1.5661% +2.7256%] (p = 0.03 < 0.05)
                        thrpt:  [−2.6532% −1.5420% −0.3001%]
                        Change within noise threshold.
Found 1 outliers among 10 measurements (10.00%)
  1 (10.00%) low mild

[philocalyst]

Verifying now 🫡

[philocalyst]

Alrighty! Okay as-was it was roughly a doubling in latency on scroll. Got some pretty simple fixes on accidental double-passes, and it's looking +15% which I think is fair. Going deeper would take deeper refactors and uh, I think we all just want this to be merged :)

[philocalyst]

OH and I added a small dependency for hashing as that was a hangup just because of the sheer amount of things that need to get hashed. gix has a hash impl but it's panick-y and annoying to work with.... could vendor/handroll if we really want to, but both are kind of ugly