Right, I see what you're saying. Yes, that won't work. I'll move this back to the bug tracker.

Hello, author I used libuv to write a network library, I used him to send UDP messages, sending 1400 bytes each time, but my speed has not been able to go up. I debugged the code and found that the recv callback is only triggered once in each loop. Because I want to reduce the CPU utilization, I called sleep (1) in idling, so the number of cycles per second is only hundreds, so the receiving speed of UDP can only be 1400 * (hundreds) per second, I want to know how to reduce the CPU frequency reasonably without affecting the network receiving speed

Version:1.19.2 Platform:windows

Libuv takes care of all that. You as an embedder just need to monitor the backend fd for readability.

I was hoping so. But as far as I can see in the source code, the function uv_poll_start () calls first uv__io_start () and then uv__handle_start (). uv__io_start () adds an watcher to the watchers queue. uv__handle_start () also does not update the epoll file descriptor. How does libuv update the uv_backend_fd file descriptor event mask after calling uv_poll_start ()?

Libuv takes care of all that. You as an embedder just need to monitor the backend fd for readability.

The idea is that you couple uv_backend_fd() with uv_backend_timeout() and then call uv_run(loop, UV_RUN_NOWAIT).

If I call uv_run (), the registered callback will probably be called. This callback can modify the event mask on any file descriptor. In order for these modifications to be applied on the file descriptor returned by uv_backend_fd (), you must call uv_run () again. But calling uv_run () again can call callbacks ..... so uv_backend_fd is potentially out of sync with the required monitoring event mask.

The idea is that you couple uv_backend_fd() with uv_backend_timeout() and then call uv_run(loop, UV_RUN_NOWAIT).

UV_RUN_ONCE should work too because you're calling uv_run() when there is definitely work to do, either processing I/O or expiring timers.

Hello. I am trying to integrate an event loop lib with another event loop. And I found out some issue. The file descriptor returned by function uv_backend_fd() is not suitable for polling. The reason for this is that some callback can add or remove event monitoring on some file descriptor, for example, using function uv_poll_start(). However, the real modification of epoll in libuv library occurs only in uv_run() function.

Maybe I misunderstand something?

• Version: commit 9c3d692b3941a2a4171629fb52af2e1029c415e8
• Platform: Linux 5.4.0-62-generic Ubuntu SMP x86_64

Most of the updates in this PR are a result of issues encountered and reported to myself by a customer. The changes are large (for one PR) because of continued work with the customer: customer reports an issue, we fix it, customer does more testing, encounters another issue, repeat - and they needed the fixes urgently.

Fixes:

1. fix a runtime crash
2. change default read-write open mode from ab+ to rb+ (to fix a locking issue when multiple processes open the same dataset)
3. fix a bug in the async functions in which an exception was thrown instead of passing an error message to the user's callback
4. handle undefined fields in the record passed to write and update (previously "undefined" would be written)
5. report an error if a field length exceeds its maxLength, instead of truncating it; also add more validation of user's arguments and I/O functions, and report failures
6. handle API calls made on a closed dataset (previously would crash)
7. add missing code that was documented with TODOs
8. remove unnecessary flags, xcode_settings and msvs_settings from binding.gyp

New functionality:

1. add 'update' and 'delete' overloaded methods that find and update/delete record(s) in a single call
2. add synchronous version for all APIs
3. enable concurrent (interleaved) dataset access:
   • use a thread per open dataset to handle all I/O (addresses https://www.ibm.com/support/knowledgecenter/SSLTBW_2.3.0/com.ibm.zos.v2r3.cbcpx01/cbc1p2274.htm)
   • parameters specific to a JS call (key, record, callback function, etc.) are now passed in each uv_work_t request, instead of shared in the VsamFile object, to support interleaved JS calls on the same dataset
4. enable key to be in any field, not just the first
5. allow 0x in a hexadecimal field
6. add support for new attribute 'minLength' in the schema JSON file

Other:

1. test: add diagnostics tests, concurrent read-access (previously would hang), write and update of undefined fields (previously wuld fail), test write and update of duplicate key, updating a key, etc. and a sync version of all async tests
2. doc: README.md updates
3. update version to 3.0.0 in package.json
4. create a wrapper WrappedVsam for the JS-C++ interface and moved the actual VSAM dataset implementation into a separate VsamFile class, to help with code maintenance

Yes, but that's not a libuv-specific question. Look up how to do cross-compiling (changing the build target) with cmake.

I have generate x64 sln for successfully using "cmake .." in windows 10. but is there any defines to generate x86 specific sln ?

thanks!

adam@g5box ~ $ node
Welcome to Node.js v14.5.0.
Type ".help" for more information.
> .helpIllegal instruction
adam@g5box ~ $ node
Welcome to Node.js v15.3.0.
Type ".help" for more information.
> .help
.break    Sometimes you get stuck, this gets you out
.clear    Alias for .break
.editor   Enter editor mode
.exit     Exit the REPL
.help     Print this help message
.load     Load JS from a file into the REPL session
.save     Save all evaluated commands in this REPL session to a file

Press Ctrl+C to abort current expression, Ctrl+D to exit the REPL
> 

Promising so far (before vs after patch). I'll do a bit more testing - is there anything simple I should try in the interpreter?