Nice work!!

Can you just ensure that the "on the wire" format is the same between your implementation and .net or c++ or one of those languages: Visual confirmation is good enough.

For reference refer to this comment I made and my observations on inconsistencies in #3144

Thanks for the review! Here's the visual confirmation of wire format compatibility between PHP and C++.

Test setup

Both tests were run inside an ubuntu-noble Docker container with the thrift compiler and C++ library built from this branch.

PHP test — writes UUID via TBinaryProtocol::writeUuid() and TCompactProtocol::writeUuid(), prints raw wire bytes:

$uuid = '550e8400-e29b-41d4-a716-446655440000';

$transport = new TMemoryBuffer();
$protocol = new TBinaryProtocol($transport);
$protocol->writeUuid($uuid);
$data = $transport->read(1024);
echo "PHP Binary:  " . bin2hex($data) . " (" . strlen($data) . " bytes)\n";

$transport2 = new TMemoryBuffer();
$protocol2 = new TCompactProtocol($transport2);
$protocol2->writeUuid($uuid);
$data2 = $transport2->read(1024);
echo "PHP Compact: " . bin2hex($data2) . " (" . strlen($data2) . " bytes)\n";

C++ test — writes the same UUID via TBinaryProtocol::writeUUID() and TCompactProtocol::writeUUID():

TUuid uuid("550e8400-e29b-41d4-a716-446655440000");

auto buf = std::make_shared<TMemoryBuffer>();
TBinaryProtocol proto(buf);
proto.writeUUID(uuid);
std::string data = buf->getBufferAsString();
printf("C++ Binary:  ");
for (unsigned char c : data) printf("%02x", c);
printf(" (%zu bytes)\n", data.size());

auto buf2 = std::make_shared<TMemoryBuffer>();
TCompactProtocol proto2(buf2);
proto2.writeUUID(uuid);
std::string data2 = buf2->getBufferAsString();
printf("C++ Compact: ");
for (unsigned char c : data2) printf("%02x", c);
printf(" (%zu bytes)\n", data2.size());

Output

PHP Binary:  550e8400e29b41d4a716446655440000 (16 bytes)
PHP Compact: 550e8400e29b41d4a716446655440000 (16 bytes)
C++ Binary:  550e8400e29b41d4a716446655440000 (16 bytes)
C++ Compact: 550e8400e29b41d4a716446655440000 (16 bytes)

Wire bytes are identical across both implementations and both protocols. UUID hex digits are written directly as 16 bytes in order — no byte swapping.

Also added UUID format validation in a follow-up commit. The writeUuid() method now validates input across all protocols (Binary, Compact, JSON), and readUuid() validates in JSON protocol (where UUID arrives as a string). Invalid UUIDs throw TProtocolException::INVALID_DATA.

This matches the approach used in the Ruby implementation (UUID.validate_uuid!).

Also added UUID format validation in a follow-up commit. The writeUuid() method now validates input across all protocols (Binary, Compact, JSON), and readUuid() validates in JSON protocol (where UUID arrives as a string). Invalid UUIDs throw TProtocolException::INVALID_DATA.

I have been thinking about this since on the NodeJS side it does validation when reading values. However I don't think that validation is something that should be happen for every field for every request, I think it adds unnecessary overhead for an application that is correct (that probably does validation outside of the message exchange).

Perhaps useful for debugging but not in operation.

@Jens-G What is your feeling around this?

Edit; see this commit I had to do on the nodets side CJCombrink@8b5576b because of the incompatibility with JAVA the node side throws away the messages since the UUID is not valid (due to the wrong byte order on the wire), this made me question 'in line' validation