Considering open source's history, you'd think its primary management tool would be open source as well. I guess it's GitHub's combination of accessible design + performant version control + lack of ads + reliability that made it the premium source for anything open source.
I'm impressed.
Their website design is beautiful, consistent, making it delightful to use everyday. They fix issues quickly, are reliable, and have top notch customer support. They try to be at the top of their game and offer features to make our work easier (while maintaining simplicity, similar to Go). You can edit posts to fix typos, and GitHub Flavored Markdown is efficient to write. They have the best API and docs I've come across.
It helps that once you're familiar with the workflow, if another project you want to make a contribution to on GitHub, it's super convenient (unlike in the past, think SourceForge and others). Even if you're not, the web flow makes it easy to press edit and go from there.
Github enterprise was a disaster with just a couple hundred users.
Under the load of just a few hundred users it died almost every day. Their support team couldn't figure it out. We had to reboot the servers constantly.
We begged them to add an option to reject force push to master. They refused. We ended up monkey patching their obscured source to add the feature, but it broke on upgrades and was hard to maintain. This is something probably every single customer of theirs wants, and it should take one engineer less than a day to implement.
They only just added side by side diff support. IMHO this is a basic, core feature for any kind of code review workflow. I'm holding my breath for proper syntax highlighting while in diff view.
We've since moved on to Atlassian Stash. While it isn't perfect, it's a dream compared to Github enterprise and a lot less $$$ to boot.
> Still, both lack a lot of basic issue tracker and code
> review features.
* The ability to mark which commits/files have been reviewed and so track progress.
* The ability to distinguish resolved and unresolved issues
* The ability to assign a reviewer based on which code is being altered in the PR.
* Useful handling of history rewriting during review.
I also think that the PR model could be implemented in a way that would make for better collaboration. At the moment, if someone submits a PR that has some issues and you want to help them fix those issues it's a huge pain because the commits typically live in their fork to which others usually don't have push access. It would be a considerable improvement if PR branches were set up in such a way that both the original submitter and the people with commit access to upstream could push to the review branch.
We host all of our private repos with Beanstalk, and use GH for our open source repos.
The fact that every comment on a review turns into a separate email is insanity, and just fills up your inbox with crap (one email for all of one person's comments on a review makes a ton more sense).
I hope they make it better, it definitely seems like a place where some TLC can go a long way.
Not trying to put you on the spot, just curious.
I think the cognitive load of using git is overrated. Yes, it could probably be simpler. No, you probably can't make simple and intuitive for everyone while retaining all the features.
Given that it's also easy to work with e.g. subversion or mercurial and a git bridge, so you can use the client of your choice with a git server, and yet no one seems to do that is an indication that git is sufficiently simple to use.
Evidence? Using my own anecdotal evidence, I strongly disagree.
I checked volume for GUI-related keywords[1] against CLI-related keywords[2].
The GUI-related keywords get ~34,000 searches per month. The CLI-related keywords get ~4,500. My methodology and data aren't perfect, but it's something at least.
Your premises are:
- People use tools that are sufficiently simple.
- People use git CLI instead of git GUIs.
- Git CLI is sufficiently simple.
I've addressed your second premise above. There isn't (and may never be) good data there, though.
[1] git gui; smartgit; git client; git tower; github gui; git frontend
[2] git command; git "command line"; git cli
So, let me restate my argument (which I suspect you may still disagree with - but hopefully this time you'll disagree with what I actually meant to claim...)
The main complaint that I have heard about git is that its command line is an incoherent inconsistent unusable mess. My opinion is that is true, it perceived to be a much bigger problem than in really is - there are lots of CLI alternatives to working with git, and their use is negligible compared to actual git CLI use. Therefore, I suspect, a perfect git CLI would not have made a big difference in adoption.
The other thing is git is incredibly flexible. Almost no one uses all the options, but much like Microsoft Word, everyone uses a different 20%. And I suspect any system with this level of flexibility is likely to be complex. Which is not to say git is at an optimum; only to say that there's a reason there's not an accepted "perfect" CLI (for git or another VCS) - inherent complexity associated with flexibility.
It even uses different words for the same thing - sometimes remove, sometimes rm sometimes -d.
Common actions often require obscure command lines, like "make a new branch" which is git checkout -b instead of just "git branch".
The reason people don't use other git frontends is that a vast majority of the help topics already out there only reference git itself. So if you want help for rebase, for example, if you're using some frontend that re-kajiggers the command line, all that help is now useless, or even worse, potentially downright incorrect.
AFAIK those usages are semantically different, and consistently so. For example, git pull actually is "repository ref" rather than "remote ref", and will go over the network to fetch the current status of ref from the given repository.
git merge, on the other hand, is simply "git merge ref"; when ref is of the form remote/branch, it will merge the current local state of remote/branch, which should be the state from the last git fetch. git merge does not access the network.
So: semantic differences, represented via different command line syntaxes. It's actually pretty reasonable, and it's pretty obvious if you've taken half an hour to understand the underlying data model of Git.
> Common actions often require obscure command lines, like "make a new branch" which is git checkout -b instead of just "git branch".
Actually, "git branch" is the command to create a new branch. "git checkout -b" is a shortcut for "create a new branch and move to that branch".
They're very close to the same syntax, and the fact that they're significantly different just shows again that the CLI is a mess. If they're so different, they should have significantly different syntax. "space versus slash" is not significantly different. I'm sure I'm not the first person to create a new branch called "upstream/master" :P
The problem with git branch versus git checkout -b is that 99% of the time, what people want to do is create a branch and switch to it. The "create a branch and don't switch to it" could just be a flag on git branch. It DEFINITELY should not be two completely different commands.
> it's pretty obvious if you've taken half an hour to understand the underlying data model of Git
This is the biggest red flag about git's CLI. I shouldn't need to know the implementation to use the UI.
It seems fairly likely that what is considered a reasonable interface or a bad interface has a high level of subjectivity. Why bother getting bogged down with such debate? If you dislike git, use something else. If git is forced upon you, write some simple wrappers which provide the interface you want for your top few most used operations, or ask for help to do so from the people forcing you to use it.
Or just hang around hacker news and complain without anything changing. It's certainly my favourite method of avoiding work :)
And then creating a branch and switching to it. Is it 'git branch -c foobar' or 'git checkout -c foobar'. Was it -c for create? why not allow it on both? And its so easy to fix these things.
This speaks to a fundamental misunderstanding of the nature of Git. You need to know the model of anything to be able to use it properly- you've mixed up Git's model with its "implementation" (probably because Git's model is so lightweight and exposes a lot of its fundamental workings, [as is right for proper usability][1]). This is like saying you shouldn't have to know what files and folders are to open a document in Word, Yes, you should.
[1]: http://en.wikipedia.org/wiki/Seven_stages_of_action#Usage_as...
Also they got a lot of clients that would have otherwise chosen code.google because Google put up barriers in terms of licences.
The money from private side has allowed them to run excellent public service without ads. And the good public service has certainly fuelled the growth on the private side.
It's the same effect.
There are only two things that are going to fix it - either something amazingly better comes along (facebook vs myspace) or the incumbent royally screws up (reddit vs digg).
Facebook, on the other hand, is one of a kind. I absolutely must have an account there if I want to do anything Facebook. And good luck getting your data out or interoperating with other social networks.
Of course, maybe I missed the memo.
https://github.com/llvm-mirror/clang
Swift hasn't been included but objective c is there.
Is the golang spec still tightly controlled by 3 wisemen in Google and no feature is allowed to add to the language without all three in total agreement?
BTW, I like 98% of their language decide choices and absolutely LOVE the compilation speed of the program.
You'll never get exceptions in Go. No one who has used go for a significant period of time wants exceptions. Error values are far superior (given the other features of the language, like multiple returns and interfaces for the error types).
You should continue to use Go, I don't think you'll miss exceptions after long. I understand the view, though... first time I saw Go I thought "No exceptions? Pass!". But I got over it, and now I've seen the error of my ways :) There's an amazing freedom that comes with knowing that random functions you call can't exit your function without your control.
If by exception you mean Java's implementation, yes. If by exception you mean not allowing programmers to ignore error values, as most Go programs seems to do at least once, no.
I think most of this could be solved easily by either a compiler option or golint rule flagging any time someone does the classic `res, _ := foo()` punt without checking the value before the end of the block or the next time that variable is assigned to. That'd satisfy the “stop repeating C's mistakes” goal most people have without requiring the other changes which true exceptions would require.
You mean like errcheck? https://github.com/kisielk/errcheck
I'm not sure if it can find spots where people intentionally throw away an error like in
val, _ := foo()
func foo() error { ... }
foo() // errcheck finds this
I've noticed most of the intention error punts in dense code where it looked like someone got tired of repeating checks and only checked the errors which happened while they were working. I'd bet a read-before-overwrite policy on _ would catch mostly infrequent environmental failures (i.e. the kind of stuff which only breaks when you're out of space, file handles, etc.).