Missing or truncated error message

[ChALkeR]

This looks close to #6456, and I thought it was just another case of that issue, but @indutny mentioned that the errors are printed from c++, so #6456 alone shouldn't have caused this, this is why I'm opening a separate issue.

A harder to reproduce example where the error is missing (it reproduces with about 10% chance for me):

for (var i = 0; i < 10000; i++) {
  console.log('HelloHelloHelloHelloHelloHelelloHelloHelloHelloHelloHelloHello ' + i);
}
a();

A simplier to reproduce example (the error gets truncated here most of the times, and sometimes is missing as in the previous example):

for (var i = 0; i < 10000; i++) {
  console.log('HelloHelloHelloHelloHelloHelelloHelloHelloHelloHelloHelloHello ' + i);
}
throw new Error(Array(100000 + 1).join('x') + '!');

[kzc]

So you're experiencing a dropped data to console with normal exiting JS? i.e., without calling process.exit()? If so, that's a different issue to #6456.

On which platform are you experiencing this issue?

[indutny]

That happens because of libc buffer size limit. In fact vfprintf returns -1.

[kzc]

src/node.cc:

    vfprintf(stderr, format, ap);

That's technically not correct without flushing libuv and node write queues to process.stderr first - and having stderr fd block.

[indutny]

@kzc this is a separate issue, IMO

[kzc]

It's a different take on the same issue actually.

$ node -v
v6.0.0

$ node -e 'for(var i=0;i<9999;++i){process.stderr.write(i + ": The quick brown fox jumps.\n")}throw new Error("Fin.");'
0: The quick brown fox jumps.
1: The quick brown fox jumps.
2: The quick brown fox jumps.
3: The quick brown fox jumps.
...
1595: The quick brown fox jumps.
1596: The quick brown fox jumps.
1597: The quick brown fox jumps.
1598[eval]:1
for(var i=0;i<9999;++i){process.stderr.write(i + ": The quick brown fox jumps.\n")}throw new Error("Fin.");
                                                                                   ^
Error: Fin.
    at [eval]:1:90
    at Object.exports.runInThisContext (vm.js:54:17)
    at Object.<anonymous> ([eval]-wrapper:6:22)
    at Module._compile (module.js:541:32)
    at node.js:328:29
    at _combinedTickCallback (internal/process/next_tick.js:67:7)
    at process._tickCallback (internal/process/next_tick.js:98:9)

vfprintf(stderr) and libuv/node writes shouldn't be mixed.

Edit: the test run above was with node 6.0.0 on Mac. Issue was reproduced with node 5.10.1 on Linux - it outputs between 2400 and 2900 rows before showing the exception on my machine.

[kzc]

Using a locally patched version of node with a flushSync method on stream it produces the expected result - outputting all rows prior to the exception.

for (var i = 0; i <= 9999; ++i) {
    process.stdout.write(i + ": stdout - The quick brown fox jumps.\n");
    process.stderr.write(i + ": stderr - The quick brown fox jumps.\n");
}
process.stdout.flushSync && process.stdout.flushSync();
process.stderr.flushSync && process.stderr.flushSync();
throw new Error("Fin.");

Edit: code sample was edited to work with the patched version of node as well as unmodified node releases. Note: this is a different patch than seen in PR #6773 which does not have flushSync method on stream (Writable actually). This patch is for demonstrative purposes only.

[kzc]

As noted above this issue is present in node 5.x on Linux (and presumably on node 4.x) so it's not a recent regression.

Not sure what's the best way to fix this mixing of different stdio mechanisms. PR #6773 will not fix it alone. It's only similar to that issue in that the event loop has ceased and stdio is unflushed.

In light of this exception/stdio issue, the documentation to process.exit is not correct:

If it is necessary to terminate the Node.js process due to an error condition, throwing an uncaught error and allowing the process to terminate accordingly is safer than calling process.exit().
https://github.com/nodejs/node/blob/master/doc/api/process.md#processexitcode

Instead of calling vfprintf(stderr,...) could v8 be re-entered at this point so that process.stderr.write() be called with the exception error string and process.exit() called to initiate the stdio flush as in #6773?

/cc @jasnell

[kzc]

Another way: invoke uv_flush_sync() on stdout and stderr uv streams, re-enter v8 just to obtain the outstanding buffered write data for stdout and stderr node streams, return to C++ and flush this data, and then call vfprintf(stderr,...).

[kzc]

Reproduced bug with node 4.x on Linux:

$ node -v
v4.4.5

$ node -p process.platform
linux

$ node -e 'for(var i=0;i<9999;++i){process.stderr.write(i+": The quick brown fox jumps.\n")}throw new Error("Fin.");'
0: The quick brown fox jumps.
1: The quick brown fox jumps.
2: The quick brown fox jumps.
...
2399: The quick brown fox jumps.
2400: The quick brown fox jumps.
2401: The quick brown fox jumps.$ 

Number of lines outputted is variable on repeated runs - 2281 to 2407. Sometimes the uncaught exception is printed, and sometimes it is not (as seen above).

[kzc]

Could someone please add a confirmed-bug tag to this issue to raise the profile of this bug?