These are great, but they're bugfixes. You'd finish them in the first few months. Then what are you going to do? Stare at an aquarium all day? Here's what I would build if I worked at github:
* Github CI - simple CI for every language, integrating with:
* Github Deploy - deploy tooling for deploying runnable artifacts or artifact combinations to either your own infrastructure, or to aws/azure/google, or:
* Github Cloud - instances/containers as a service
and the 'fork' button would be a pulldown that would include the options 'fork', 'fork and test', 'fork, test and deploy'. Now that's a nice little 3-4 year career.
Non of the items in the author's list is a bugfix. They all are enhancements. A bugfix would fix a bug, which means something would not work as it is supposed to, fail, and that is not the case here.
Arguing semantics, maybe. But I think it is important to distinguish here. Polishing issues or enhancements or whatever you call it are things that do not need to be adressed, but they can. But bugs needs to be fixed. Choosing between what to do now is choosing between maintenance work and feature enhancement work, plumbing and UX.
I think it would be great if Github would hire and let him build these things, does not matter if he finishes in a month.
Of all the things I wish Github had, a mailing list service would be really amazing, and the kind of thing that wouldn't be done in the first few months. Sourceforge offers atrocious mailman-based email, larger foss projects on github suffer :(
If it's reporting issues, discussion solutions, and reviewing patches, there are Issues and Pull Requests.
If it's for general discussions about the "future of the project" or bigger-picture topics, I'd use Discourse. It's a really nice way to organize discussions that are one step removed from code-related issues.
If it's for things where you need quick or real time feedback, there's IRC, Gitter, and Slack, all great options for messaging/chat.
Isn't this a bit like saying: if it's for issue tracking, there's Bugzilla/trac/issuetracker/phabric, or if you need a wiki there's moinmon/mediawiki, or if you need a blog there's ghost/wordpress ?
More to the point: IRC (out-of-the-box) doesn't do archiving/search. Slack isn't self-host (which isn't an issue with people using github -- but it does introduce another vendor). Using external services forces you to maintain group membership, user-meta-data either in different user-databases, or via some form of federation.
No longer is removing a user/ssh-key from a github project enough to plug a hole in case of a hacked account.
Discourse isn't (that) bad -- but mailinglists are a lot better IMNHO. If you have a half-decent email program, like mutt, or even (al)pine.
At any rate, the ability to work via email (get bugs via email, close bugs via email) leverage what email is good at: off-line+synchronization. Which is one of the things git is good for. You know, distributed work.
Not sure about authorization and group membership, channel authorization etc -- I assume you'll need to manage slack authorization with slack, and github authorization on the github side.
> If it's for general discussions about the "future of the project" or bigger-picture topics, I'd use Discourse. It's a really nice way to organize discussions that are one step removed from code-related issues.
This is what is appealing and Github offering a builtin-to-github solution for it, with linking to Github accounts, autoformatting etc (like in issues) and so on. There is so much potential.
That's an interesting idea (one of my big blockers for migrating my projects off SourceForge is what to do with the mailing lists).
The issue UI is a little forum-like but is pretty clunky, and, of course, doesn't interact with email very well. Plus you need people to manually tag each issue with 'discussion'.
Are there any UI tweaks that could simplify this? For example, a way to provide a 'new issue' link with default labels?
(Also, is there a good way to import data as issues? I have a tonne of mbox archives that need to go somewhere.)
No, the blocker is setting up mailing lists with a forum-like interface (cf. discourse), but integrated to github (uses the github account) and without actually having to do the setup. Also, without polluting issues.
Actually opening issues via email would be nice to. Trac isn't perfect, but I remember email2trac[1] quite fondly. Combined with email notifications on change, it made it easy to forward bug/error-reports to a trac instance, and automatically create an issue. Or just let support@ go straight to trac.
I still think having an email interface (as in: in addition to a REST API) is very useful. Having a web interface as well isn't bad either.
Continuous Integration. A good example of that service would be something like http://travis-ci.org or https://jenkins-ci.org. Basically it's a service that runs some predefined task (usually your test suites) at important points: when something is pushed to or merged into master, when a Pull Request is opened, etc...
Great minds... I was talking about this the other day with the other engineers at work, after learning about GitHub's most recent monstrous funding round. It seems like CI/a deployment service and something that competes in the crowded PaaS field is the next logical step.
AWS enables us to do some pretty slick things, but OpsWorks is a giant bag of suck that frequently finds a way to drain my time troubleshooting why something crazy or random has happened on the production server--behaviors I never saw in my previous role running a bare-metal LAMP stack for 4 years. A tighter integration between our GitHub repo and a testing/build server... Swoon.
A github CI would be awesome! I imagine it would get very costly for them though. I love Travis-CI and the easy integration, but it's too expensive for my non-open source projects.
All these are nice, but are not must-haves as they can be built outside of GitHub. The show-stopper for me is granular permissions - coming from the Gitolite [0] world, where I can permission even branches and do all kinds of checks and balances on the server, GitHub is a joke for a not-so-large enterprise! Another thing is the Wiki - why pages are based on Git and the Wiki is not?!
At some point in the last year or so Github rolled out a new search engine which drastically reduced it's usefulness. Any moderately complex search query now has all of the modifiers and key bits stripped out making your search results unnecessarily cluttered and somewhat useless. I consider it to be one of the worst parts of the site these days.
> While that's true, I think the discussion in that issue has gone more towards HTTPS support for custom domains.
A huge number of those "+1 for HTTPS on custom domains" don't seem to understand/appreciate the difference in providing HTTPS on the *.github.io (one wildcard cert) vs. HTTPS support for custom domains.
The latter would require an interface (UX being key here) and storage for uploading your own domain certificates to GitHub, which is nothing like any other part of GitHub right now. I also presume that most of these "+1's" would want this service to remain free.
You could always use CloudFlare directly for that... I would presume that their underlying connection to Github is very close and reliable in terms of risk for MITM.
It seems like some of these (perhaps all of them other than fixing the "+1" problem) can be done via API clients instead of via GitHub itself implementing it. A third-party website can do wiki searches. A separate service can send you email notifications, and you can turn off built-in emails.
Even the +1 thing might be solvable with a bot that edits each bug report to add a clickable +1 badge (a la the CI build-passing badges) at the top; all you need is to give the bot ownership of your repository. The badge can display the current +n count, and the service can give you a sorted list of open issues by +n. For bonus points, have the bot also harvest and remove comments that consist of just "+1" or ":+1".
This is GitHub after all; why don't we build stuff ourselves instead of waiting for a centralized closed-source company to decide they care about our features?
>This is GitHub after all; why don't we build stuff ourselves instead of waiting for a centralized closed-source company to decide they care about our features?
Why would I build something for a centralized closed-source company?
Because then the only thing you're relying on the centralized closed-source company for is actual git hosting. This is a boring problem that dozens of other services solve well, and thousands of other services (like S3) solve poorly in a pinch. If you can migrate the interesting part of GitHub -- issue tracking, PRs, wikis, etc. -- to another provider, and GitHub just holds your data, then you're no longer locked in.
If you don't build it and you wait for the centralized closed-source company to, then you're putting your project even more in their hands.
You could build the features for GitLab. We'll roll them out on GitLab.com that you can use for free. We can import repo's and issues already and are working on PR's and wiki's.
Migrating issue tracking and other stuff out of Github seems like a good idea if you want to reduce dependency, building a bot on top of Github issues - what you implied in your parent post - not so much.
ZenHub was the name I was trying to think of, thanks, but it seems like a thing everyone needs to opt into to make good use of (I think? in that it's a browser extension?). The idea I had is more along the lines of the CI badges, which don't require any action on the end-user's part. It's just an image inside a link, so it works in every browser already.
If ZenHub supports making its features available to OSS projects without needing every contributor to install the extension, that's much more compelling, and also totally not obvious from their website.
>it seems like a thing everyone needs to opt into to make good use of
Yep! Which is why I gave up on trying to use it. I thought it would gain major traction given how useful it is - but I felt like the only person using it. So I eventually removed the extension.
>Not affiliated, used it, but without others using it it's largely...well.. useless. :)
The only thing I really, really want is paginated diffs for pull requests. C'mon. If I have 300 files, don't show me the first 15 and then say "well, this is too big, you'll just have to imagine what the rest of the diff looks like." Does Google return the first page and then say "the rest of the internet isn't available to you because we don't want to paginate things"? No, of course not. Just freaking paginate long diffs.
If today you see that a major repository has a fork with a lot of commits, you cannot know without looking through THEM ALL if they are just typo fixes or if the fork is really changing/improving things over the main repo.
One way that has served me as being a pretty good metric of what's what when it comes to forks looking at the number of stars the parent and child repos have. If a fork has started to usurp it's parent, it generally has a fair deal of its own stars.
100% agree. I wish there was some way to also promote a fork, other than changing the readme to point to the new fork, which also means mucking up the history of the project.
The biggest thing I want from GitHub is the ability to search commits. GitHub are pretty good at search (their issue search is fantastic, and their code search is decent enough) - but the data I most want to search are the commit messages in my repo.
I understand this will be a huge amount of data (they must have hundreds of millions if not billions of commits by now). I'd be perfectly happy if this was only available for paid repositories.
The ability to then further facet and filter searches by author, file, directory etc would be unbelievably useful.
Oh, and while we're talking pony requests... I'd dearly love to be able to use gists within the scope of an organization. It would be fantastic to be able to share code snippets that were only visible to other organization members (and searchable and so on).
By far the biggest thing I wish github would do is change the text entry beside the file names. Right now, it shows the comment of the last push of edits to that file. That's great for members of the project.
However, for non-members it would be far more useful to know what the file is. Here, enabling a one-line description would be hugely helpful.
Since GH knows if you're a member of the project it can show you the right text--description or latest update. And presumably, a button would allow you to see the other text if you needed it.
If you don't know where to look it can take a surprisingly long time to find things like: https://github.com/torvalds/linux/blob/master/Documentation/.... It is obvious once you know, but unless you know are in the habit of checking all the 'likely places' such a doc would be you are going to be lost. While one could allow annotations on top of files in github it would completely fragment real and important documentation efforts and leave information scattered in 10 different locations that no one could pull back together.
Lots of things. Github ships features like crazy. Lots of the features I use every single day were only released in the last few months. On a scale of years, all of their desktop clients, which are a huge experience improvement for new users.
I see people here posting issues of their own, but I'm amazed nobody has yet mentioned the lack of IPv6.
I've setup a few machines with IPv6-only network connections, and it makes pulling my dotfiles from github a pain since I need an IPv4-proxy/tunnel to access github.com
Great list, we're working on making something that makes +1's less noisey. If you want the possibility to contribute to the tools you use daily consider using our GitLab. It is open source and more than 800 people have contributed already.
I think GitLab supports it https://gitlab.com/gitlab-org/omnibus-gitlab/issues/412 but we have not had the time to set it up for GitLab.com yet. We'll be moving our infrastructure to Azure soon so it is not a priority for us now.
code is read way more often than written. and on github website all you can do is read it. why the viewports don't take up the entire screen width i will never understand.
It is long lines which damage readability, and that is the fault of the gonk who wrote the code. If anything, GH's fixed-width design encourages shorter lines and so improves readability.
I don't use it often, but I have a Javascript bookmark [0] that I use in Chrome when the 100-character width is excessively abused. Over time, I found I dislike using it enough that I now try to limit my lines to 100 characters in my editor, so that my github diff presentation is easier to read.
I'm sure there's an easier way to do this. I'm pretty certain I cribbed the method of setting a style from a StackOverflow article or other thing found online, but that was three years ago so I do not recall where it was. (Sorry.) I think all I did was poke around the Chrome debugger until I found the styles I needed.
My biggest pet-peeve about GitHub is its issue tracker system. It's great for one-off issues or tracking a pull-request, but beyond that its design is super unwieldy. Such a missed opportunity to crush JIRA, et. al.
It adds these features to improve the GitHub Issue tracker directly into GitHub.com, allowing you to:
- Display issues on a task board by priority and progress
- Attach any file type to an issue
- Add estimates and view burndown charts / metrics attached to them
- +1 issues to avoid excess comments
- Improved search functionality
- Link repos together to view issues that span multiple projects
Without having to context switch out to a separate issue tracker or UI
----
Disclosure - I work at ZenHub, and use it with GitHub all day :D
Generally the UI is just poorly designed and unfocused. One specific thing: You can't mark them as approved in any way and doing something like +1 in a comment somewhere doesn't count.
Personally, I'd like a nice GUI to pick and squash commits with new commit messages against them on creating a pull request. I don't enjoy using the CLI for this, and it would be nice to thread thoughts on the solution into the commits themselves rather than explaining solely in the PR description.
I second this, also I'd like to see an overview of the (files with) merge conflicts if any instead of a message saying the PR can't be merged. Fixing the merge conflicts might be another thing in the web interface would be cool but I guess I'll stick to the CLI for this.
- No ability to store the tab size setting. While appending '?ts=2' to the url is ok, it's not convenient and usable on a daily basis. The same for omitting whitespace changes in a diff / pr ('?ws=1').
- No ability to step through history on a single file while in blob view, or blame view. In blame view, the commit link goes to the commit, not the blame of the file at that commit.
- Agree with OP that notifications need to be grouped / summarized better. Perhaps expandable summary items per repo per type. This is mostly useful for private, work-related repos. Pulse and notifications are basically the same thing.
- Stars cannot be tagged. This makes managing hundreds of stars difficult. There's apps like Astral, but even those are lacking.
Those are interesting suggestions, but my guess is that they are probably focusing on features to make github more enterprise friendly. If a16z invests $100 million in github, it doesn't seem like streamlining emails will have the ROI that investors expect.
Maybe github is focusing on creating a full software development lifecycle management (ALM) in the cloud. (Like Microsoft Team Foundation Server and JIRA.) A dashboard for sprints, defect fixes, issues tracking, etc. That's the type of "enterprisey" thing that attracts more business subscriptions. They use the storage of sourcecode repositories to open doors to other sw development related business.
These are just my guesses. I haven't seen any explicit roadmap from github.
I'm actually hoping they bring something into GH for business/enterprise that is better, or as good as TFS or JIRA... I like TFS, but non-VS projects are a pain... and JIRA is just well, not pleasant at all.
Once they have some sort of better issue tracking system, then migration tooling will become another big issue in that space.. getting from SVN+JIRA or TFS itself would be really useful to a lot of orgs, willing to pay. Something between TFS and Axosoft's OnTime would be pretty nice as an addition to GH.
Another area with a lot of potential would be a CI/CD toolset that uses your own cloud.. EX: you use api keys for Azure, GCE, AWS, DO etc, and then you can setup testing/limits and Github could just manage the provisioning of the test servers, and ease setup/deployment of test/stage/prod classes of servers for multiple languages.
GH wouldn't have to manage a huge infrastructure itself, just build/sell tooling that extends it's current product line.
Wiki search is #1 on my wish list for Github. I see others talking about cloning the wiki repo locally and searching it but for big projects this would be the equivalent of downloading something like the Python docs and searching them. It could be done but it would be a massive pain in the ass.
I have written most of the documentation for the project I contribute to the most and I organized the wiki like a two level tree where a top level page has links to category pages and each page on the wiki is linked to from one of those category pages. Then there are links interspersed among the pages like a normal wiki. This structure works okay but I know it can be better because sometimes I cannot even find information that I produced!
I feel one could have more ambitious goals. E.g. Could Github build a powerful developers careers site based on all the data they have? Could they provide more tools to help people program? (Could they automatically create programs? https://news.ycombinator.com/item?id=9973088)
* Push conflicted branches so developers in a distributed team can work together to resolve conflicts.
* A git-powered version of Rake, which runs only those tests that need to be run based on the git history since the last run.
* A tool to identify what code changes caused a particular test to fail, based on the above.
* Language syntax detection for smarter diffs, improving the display when blocks of code are moved around or indented/outdented.
* Language linters built in, to detect when a change is introducing a syntax error. Same for coding style.
I'd also move the "Close Pull Request" button a little farther away from the "Comment" button, and make it possible to add comments to a diff when you've hidden whitespace differences (with `&w=1`... I'd also make it a bit more obvious that you could do that).
Well, it is kind-of possible for Ruby/Rails. I half-built such a system once. Here's what you do:
* Gather coverage information when running tests or compiling assets.
* Use that to generate Rakefiles for running tests / compiling assets that have dependencies on the source files they would touch.
* Run the tests once, recording information about the git status of each dependency.
* When the git status of a file changes (i.e. you've just fetched some changes), mark the dependency as dirty
Even better, it you're gathering coverage data and know the test status for each commit, then not only can you show the commit that broke the build, you can show the particular part of the diff that broke the build.
> The suggested alternative is to "Subscribe" to an issue for updates, which is what I've started doing to issues that I want to show my support for. My support, however, isn't visible to anyone else.
Though following the resolution of an issue usually means I'd like for it to be resolved, the converse isn't true: I've hit thousands of issues over the years but most of the time I've been able to work around them and haven't hit it since, resolving the issue would be nice for people coming after me, but I don't really care to be spammed by its status updates.
So yes, "subscribe" could count as a +1, but no "subscribe" should not be the only way to "+1" an issue.
I always wished GitHub had a chat room per repo. Fully integrated with github features, like mentions, issues, pull requests. Maybe even hubot builtin.
How about making pull requests less centralized? Right now, if I file a PR, only the project admins or I can modify/rebase it. I'd like to give other (non-admin) collaborators permission to work on it too.
Or, maybe it could use a forking model? Let anyone fork their own PR from mine, make those forks prominently visible in my base PR, and make it easy for me to merge their edits back in.
If you want anything less centralized, then why are you using Github, a centralized interface to and storage of a previously inherently de-centralized system, git?
I reported this a while back, so it's in their backlog somewhere but obviously not super-important:
I'd like to see an option to turn off fork notifications for org projects without turning off all org notifications. As-is, I get an email for every dev forking every project, sometimes more than once due to lack of git knowledge.
I would make the search results deduplicated. It happens to me a lot that I search for something and the first 10 pages of search results are from the same 5 projects, but from different locations within the single projects. I usually lose patience sometime around page 5.
It's very common, at least on my team, to create pull requests for the wrong target branch. I don't see why GitHub doesn't allow you to fix it before any comments are posted. This would save some useful time.
For that matter, on the issues page, and labels/markers for new issues vs. old ones with a "me too" comment. or for that matter issues that you've commented on.
I follow a couple dozen projects and after a couple days, my GH notifications page gets insane. Making issues more usable would be a big win imho.
Also, a "priority" follows would be nice too.. so your org/project issues can be viewed separately from your follows. Seeing adoption of something closer to ontime/tfs/jira would be nice too.
To be frank, it is pretty much ok for the +1 to be annoying: it's often what they are meant to. Thus, if a vote button were to be implemented, it would not be used all the time.
I would really like to have the ability to Edit commit messages using the web interface. I'm often stuck with a poorly crafted message and an edit function would help that.
Ugh, that should be an opt-in thing. I've had trouble with Trac parsing commit messages as trac-wiki syntax, which is generally a great idea. The problem is occasionally you'll use markup characters in a semantically meaningful way, e.g. "Switch from `foo` to $(foo)", and rendering the markup will make the commit message confusing to read.
I'd be much more enthusiastic if there was some magical way for the local git command line to also support Markdown rendering/preview.
I'd argue that's still a bit of a pain compared to a fulltext search in the browser. I can already search the rest of the repo and issues that way, so why not the wiki?
* Github CI - simple CI for every language, integrating with:
* Github Artifacts - repository for versioned deployable build packages (binaries, tar.gz files, ios builds, android builds...), integrating with:
* Github Deploy - deploy tooling for deploying runnable artifacts or artifact combinations to either your own infrastructure, or to aws/azure/google, or:
* Github Cloud - instances/containers as a service
and the 'fork' button would be a pulldown that would include the options 'fork', 'fork and test', 'fork, test and deploy'. Now that's a nice little 3-4 year career.