Binder refactor #1989
Binder refactor #1989
Conversation
nateberkopec
changed the title
lib/puma/binder.rb
Outdated
| when UNIXServer | ||
| ["unix", i.local_address] | ||
| when Puma::MiniSSL::Server | ||
| ["ssl", i.to_io.local_address] |
There was a problem hiding this comment.
This has the advantage of logging port 0 binds with SSL correctly too.
|
@MSP-Greg thoughts? |
|
Oh, on windows, UNIXServer isn't even defined? |
|
I need to think about how this affects inherited listeners, and whether or not I should be iterating through |
nateberkopec
force-pushed
the
MSP-Greg-unix-and-tcp-listeners
branch
from
eabef54
to
f892e80
Compare
|
I don't think it matters whether or not I'm iterating through listeners or ios. Unfortunately, those are supposed to be the same data AFAICT. I think we should probably just have one collection in binders (merging ios and listeners), a collection of a new value object (call it Binding?). That would make a lot of stuff a lot nicer. I don't really want to do that refactoring right now but I think I have to in order to log inherited sockets once rather than twice (which this will now do). |
nateberkopec
force-pushed
the
MSP-Greg-unix-and-tcp-listeners
branch
from
58e4c69
to
4da434a
Compare
|
This is ending up a lot more intense than something that can just fix that bug and go into a patch release. Once I finish this I'll walk it back to something that can, maybe even just Greg's original fix + test. |
nateberkopec
force-pushed
the
MSP-Greg-unix-and-tcp-listeners
branch
from
6b84cd1
to
0ab4094
Compare
nateberkopec
force-pushed
the
MSP-Greg-unix-and-tcp-listeners
branch
from
0ab4094
to
cc03819
Compare
nateberkopec
force-pushed
the
MSP-Greg-unix-and-tcp-listeners
branch
4 times, most recently
from
6d102f5
to
c35cccf
Compare
nateberkopec
force-pushed
the
MSP-Greg-unix-and-tcp-listeners
branch
from
c35cccf
to
60d7fe7
Compare
Co-authored-by: MSP-Greg <MSP-Greg@users.noreply.github.com>
|
I think it's time for me to stop the spike for now and pull back a few of these things into master. |
|
As to the emoji I added, it reminded me of how "I want to fix a few issues" often turns into a lot more. At least that's been the case with me...
I think the current code only loads listeners in the parse code. ios is loaded in the add/inherit code. So, control url's aren't in listeners, and any sockets created via re-entrant calls in the add/inherit code are also not. I think... |
This commit extracts the `MiniSSL::Context` creation into its own `MiniSSL::ContextBuilder` class along the same lines as in [puma#1989]. This will allow us to reuse this code for adding SSL support to the control app (issue [puma#2015]). Since we will need the `MiniSSL` require and check in both places, I moved that into the `ContextBuilder` class as well. [puma#1989]: puma#1989 [puma#2015]: puma#2015
* Extract class for building SSL context This commit extracts the `MiniSSL::Context` creation into its own `MiniSSL::ContextBuilder` class along the same lines as in [#1989]. This will allow us to reuse this code for adding SSL support to the control app (issue [#2015]). Since we will need the `MiniSSL` require and check in both places, I moved that into the `ContextBuilder` class as well. [#1989]: #1989 [#2015]: #2015 * Add SSL support for the control app This starts to address [#2015]. I think we will need to add SSL support to the control cli as well. [#2015]: #2015
|
Closing as unmergeable, but I do want to revisit the ideas in here very soon. |
Doing an inheritance model could probably help me take a lot of the complexity out of Binder#parse and the various inherit_ and add_ listener methods...