[Libwebsockets] Deprecating Github 2018 The Man edition(R)
andy at warmcat.com
Mon Jun 11 08:42:53 CEST 2018
On 06/05/2018 11:30 AM, Brice Hamon wrote:
> I was thinking just that today Andy.
> I share all what you said and I am in the same boat..almost.
> Seems like self hosted is the safest route, minus the pain.
Yes I think many people got a wakeup call with this.
It took a while because I had to find a bug around closing cgi dup2
stdin handles correctly, so EOF can appear two processes down inside git
for when it's serving https:// cloning. That's a few lines but
understanding what the problem was took several days, replacing git with
a locally-built one, debugging via stderr from the cgi etc. It's fixed
Other CGI pay attention to CONTENT_LENGTH but git ignores it and just
does a blocking read on stdin until it gets an EOF, >8K or times out.
From now on the primary place to get lws is officially:
git clone https://libwebsockets.org/repo/libwebsockets
github links are scrubbed from https://libwebsockets.org and the
The next problem is figure out what to do about the "forge" aspects...
cgit gets the presentation of the repo quite far but I can't find any
way to enable inline display of README.md in tree view, which was a good
innovation from github. It does it for the top level on the about tab well
... but the inner directory README.md -s are not rendered by default.
Being able to deploy different "forge" contexts has its own uses for me
so I'll look down that road first.
> Thanks again Andy for your hard contribution
> On Mon, Jun 4, 2018 at 8:27 PM, Andy Green <andy at warmcat.com
> <mailto:andy at warmcat.com>> wrote:
> Hi -
> About the github purchase by microsoft... I have spent getting on
> for two decades trying to keep microsoft's clammy little hand off my
> leg, but it keeps reappearing there.
> So I want to discuss ideas for removing lws (and my other stuff
> there) from github, just leaving either a mirror or a README.md
> pointing to the replacement.
> I considered gitlab.io <http://gitlab.io>, however while it's good,
> it doesn't permanently fix the same "relying on proprietary
> services" hole that created this mess... it's just kicking the can
> down the road.
> Looking at what infrastructure lws uses with github rejected, we have:
> libwebsockets.org <http://libwebsockets.org> the server: US$10/mo
> lwsws : FOSS
> git-daemon / cgit : FOSS
> mailman : FOSS
> appveyor (windows CI) : PROPRIETARY | $0
> travis (mac + Linux CI) : PROPRIETARY | $0
> coverity (static analysis) : PROPRIETARY | $0
> So if we put the CI stuff to one side (the free OSS Coverity has
> usage limits, I use it mainly when leading up to a release) the
> original selfhosted stuff will support ongoing development OK. In a
> pinch I can set up Jenkins and implement the CI builds and tests for
> Linux selfhosted as well (plus or minus some pain).
> <https://libwebsockets.org/git/libwebsockets> has always been kept
> in sync with lws git, it needs updating with a cgi implementing to
> support https which hopefully won't be too hard.
> So basically the choices seem to be selfhost (https://gogs.io/ looks
> possible to implement without it being a fulltime job), or just
> don't have forge and go back to selfhosted git + mailing list.
> If you have ideas or comments let me know.
> Libwebsockets mailing list
> Libwebsockets at ml.libwebsockets.org
> <mailto:Libwebsockets at ml.libwebsockets.org>
More information about the Libwebsockets