authorgravatar for mlugg@mlugg.co.ukMatthew Lugg <mlugg@mlugg.co.uk> 2024-11-27 00:10:33+00:00
committergravatar for mlugg@mlugg.co.ukMatthew Lugg <mlugg@mlugg.co.uk> 2024-11-27 19:35:31+00:00
log1a99c99ee99fa3ca43935fe7d7d0865145b82d4e
tree977a570f2bf72ddfaa6fe7f8d754161c5e4581ad
parent3ce6de87657e67da358e9c7869a87c61cc9f1305

std.Build: gracefully handle child stdin closing when running tests

We have deduced that it seems the sporadic BrokenPipe failures happening on the CI runners (e.g. https://github.com/ziglang/zig/actions/runs/12035916948/job/33555963190) are likely caused by the test runner's stdin pipe abnormally closing, likely due to the process crashing. Here, we introduce error handling for this case, so that if these writes fail, the step is marked as failed correctly, and we still collect the child's stderr to report. This won't fix the CI issues, but it should promote them to proper error messages including child stderr, which -- at least in theory -- should allow us to ultimately track down where the errors come from. Note that this change is desirable regardless of bugs in the test runner or similar, since the child process could terminate abnormally for any number of reasons (e.g. a crashing test), and such cases should be correctly reported by the build runner.

1 files changed, 33 insertions(+), 10 deletions(-)

lib/std/Build/Step/Run.zig+33-10
...@@ -1413,12 +1413,23 @@ fn evalZigTest(...@@ -1413,12 +1413,23 @@ fn evalZigTest(
1413 });1413 });
1414 defer poller.deinit();1414 defer poller.deinit();
14151415
1416 if (fuzz_context) |fuzz| {1416 // If this is `true`, we avoid ever entering the polling loop below, because the stdin pipe has
1417 try sendRunTestMessage(child.stdin.?, .start_fuzzing, fuzz.unit_test_index);1417 // somehow already closed; instead, we go straight to capturing stderr in case it has anything
1418 } else {1418 // useful.
1419 const first_write_failed = if (fuzz_context) |fuzz| failed: {
1420 sendRunTestMessage(child.stdin.?, .start_fuzzing, fuzz.unit_test_index) catch |err| {
1421 try run.step.addError("unable to write stdin: {s}", .{@errorName(err)});
1422 break :failed true;
1423 };
1424 break :failed false;
1425 } else failed: {
1419 run.fuzz_tests.clearRetainingCapacity();1426 run.fuzz_tests.clearRetainingCapacity();
1420 try sendMessage(child.stdin.?, .query_test_metadata);1427 sendMessage(child.stdin.?, .query_test_metadata) catch |err| {
1421 }1428 try run.step.addError("unable to write stdin: {s}", .{@errorName(err)});
1429 break :failed true;
1430 };
1431 break :failed false;
1432 };
14221433
1423 const Header = std.zig.Server.Message.Header;1434 const Header = std.zig.Server.Message.Header;
14241435
...@@ -1437,13 +1448,13 @@ fn evalZigTest(...@@ -1437,13 +1448,13 @@ fn evalZigTest(
1437 var sub_prog_node: ?std.Progress.Node = null;1448 var sub_prog_node: ?std.Progress.Node = null;
1438 defer if (sub_prog_node) |n| n.end();1449 defer if (sub_prog_node) |n| n.end();
14391450
1440 poll: while (true) {1451 const any_write_failed = first_write_failed or poll: while (true) {
1441 while (stdout.readableLength() < @sizeOf(Header)) {1452 while (stdout.readableLength() < @sizeOf(Header)) {
1442 if (!(try poller.poll())) break :poll;1453 if (!(try poller.poll())) break :poll false;
1443 }1454 }
1444 const header = stdout.reader().readStruct(Header) catch unreachable;1455 const header = stdout.reader().readStruct(Header) catch unreachable;
1445 while (stdout.readableLength() < header.bytes_len) {1456 while (stdout.readableLength() < header.bytes_len) {
1446 if (!(try poller.poll())) break :poll;1457 if (!(try poller.poll())) break :poll false;
1447 }1458 }
1448 const body = stdout.readableSliceOfLen(header.bytes_len);1459 const body = stdout.readableSliceOfLen(header.bytes_len);
14491460
...@@ -1483,7 +1494,10 @@ fn evalZigTest(...@@ -1483,7 +1494,10 @@ fn evalZigTest(
1483 .prog_node = prog_node,1494 .prog_node = prog_node,
1484 };1495 };
14851496
1486 try requestNextTest(child.stdin.?, &metadata.?, &sub_prog_node);1497 requestNextTest(child.stdin.?, &metadata.?, &sub_prog_node) catch |err| {
1498 try run.step.addError("unable to write stdin: {s}", .{@errorName(err)});
1499 break :poll true;
1500 };
1487 },1501 },
1488 .test_results => {1502 .test_results => {
1489 assert(fuzz_context == null);1503 assert(fuzz_context == null);
...@@ -1518,7 +1532,10 @@ fn evalZigTest(...@@ -1518,7 +1532,10 @@ fn evalZigTest(
1518 }1532 }
1519 }1533 }
15201534
1521 try requestNextTest(child.stdin.?, &metadata.?, &sub_prog_node);1535 requestNextTest(child.stdin.?, &metadata.?, &sub_prog_node) catch |err| {
1536 try run.step.addError("unable to write stdin: {s}", .{@errorName(err)});
1537 break :poll true;
1538 };
1522 },1539 },
1523 .coverage_id => {1540 .coverage_id => {
1524 const web_server = fuzz_context.?.web_server;1541 const web_server = fuzz_context.?.web_server;
...@@ -1552,6 +1569,12 @@ fn evalZigTest(...@@ -1552,6 +1569,12 @@ fn evalZigTest(
1552 }1569 }
15531570
1554 stdout.discard(body.len);1571 stdout.discard(body.len);
1572 };
1573
1574 if (any_write_failed) {
1575 // The compiler unexpectedly closed stdin; something is very wrong and has probably crashed.
1576 // We want to make sure we've captured all of stderr so that it's logged below.
1577 while (try poller.poll()) {}
1555 }1578 }
15561579
1557 if (stderr.readableLength() > 0) {1580 if (stderr.readableLength() > 0) {