Set sync or flush_on_newline for standard I/O on Windows.
#9207
Conversation
|
Yes, it is better to move the code as you described. |
kubo
force-pushed
the
standard-handles-on-win32
branch
from
151a184
to
2e1a236
Compare
RX14
added
the
platform:windows
If starndard I/O is a handle to the console, set `sync` to true. Otherwise, set `flush_on_newline` to true as described in STDOUT and STDERR docs.
kubo
force-pushed
the
standard-handles-on-win32
branch
from
2e1a236
to
e43943a
Compare
|
Add a comment as requested by @straight-shoota here. |
|
This should be merged as separate commits btw. |
Could you explain more specific? I cannot understand your request. |
|
Sorry for confusion :( |
|
Sorry, we can't do that any more, @waj decided everything had to be squashed. |
|
Yeah, how lovely... |
|
Please let's not say X decided something. Many of us agree that squashing is fine because you can always go to the PR to see the commit history. @oprypin do you think it would be fine to squash this anyway? In my experience whenever I ended up using blame and in a single commit, I navigated to the PR anyway to understand the overall context. |
|
I would be fine with always squashing, and announcing that that is the case, then people won't have any reason to force-push into the PR branch. Otherwise we're getting the worst of both worlds. |
|
I've had an idea for a solution which would make everyone happy for a few years, but unfortunately it requires writing code and starting a new project is hard to justify... I apologise for being ratty this morning, I think I might need to take a break from Crystal... |
|
I think that's another benefit of squashing: PR authors don't need to care about keeping a clean commit history. That should speed up contributions. |
This PR sets
syncorflush_on_newlinefor standard I/O on Windows as on Unix.This uses GetConsoleMode, which succeeds only when the handle is to the console.
I tested it by the code in #9149 (comment). Without this PR, the benchmark results are displayed on process exit. With this PR, they are displayed as on Unix.
I want to hear opinions. Is it better to move win32 and unix code in
IO::FileDescriptor.from_stdiotocrystal/system/win32/file_descriptor.crandcrystal/system/unix/file_descriptor.crrespectively?