Reopen stdio streams when run from Drip, to pick up its FIFOs. #4010
+272
−145
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This fixes #2690, mostly.
Previous situation
On startup, we assume stdio will be at the usual 0,1,2 values descriptors, and we use those directly to open native stdio streams for Ruby. This conflicts with Drip background processes, since after the "drip main" has run, it switches System.in/out/err to point at FIFO file streams that are used to proxy the client process. If we assume stdio is always at 0,1,2, we wire up the wrong descriptors.
First fix
My first fix was to just skip the native stream logic and use our tried-and-true unwrapping logic to take System.in/out/err and unwrap them down to their innermost File*Stream. When running under Drip, one of its Switchable streams will be in the way, and we will fall back on using wrapped-stream logic for stdio. That fixes this issue at least well enough for stdio to function.
Unfortunately this fix has a problem: by not unwrapping all the way to the FileChannel, we lose some interactivity at a console. Granted, we already lose some interactivity due to Drip's FIFO being in the way, but on top of that we also have Java's stream buffering interfering with interactive use.
Additional fix
I continued with the above fix and added smarts to further unwrap any Switchable streams. After much fussing around, I have the logic properly getting Drip's FIFO file descriptors for stdio, and fully unwrapping them.
Unfortunately, by default when Drip boots JRuby, it attempts to use our org.jruby.main.DripMain to start up a background JRuby instance. This instance sees the stdio streams before they have been switched to the FIFOs, and so that prebooted instance has stdio pointing at the wrong descriptors.
We have addressed this for other Drip laziness by adding Ruby.reinitialize, which re-inits anything we need to defer until the process "really" starts. For example, ARGV is re-defined to point at the actual arguments passed in by the Drip client.
I also did this work so that the streams would be reinitialized, by replacing the existing descriptors in the stdio streams with the new ones from Drip. However, any code that grabs the Ruby stdio file descriptors before the "true" process start will continue to point at the old, incorrect descriptors.
Possible solutions
$stdin
and friends...they just use top-level IO functions that write to those streams. But there's always troublemakers...A side effect from my changes here is that stdio.fileno will now always reflect the actual fileno. Previously I forced stdio streams to always report their fileno in the 0,1,2 group, which doesn't make sense if they're not actually using those fileno.