fix: better handling of invalid headers

[dmehala]

Description

When the tracer encounters an invalid propagation header value, logging the list of headers and their values is not handled properly.

Details

For each key and value in the list of headers, a nlohmann::json instance is created with the following code:

stream << nlohmann::json(it->first + ": " + it->second);

It is possible that an input for any of the headers inspected by the tracer exhibits unexpected behavior when the JSON library attempts to parse or dump the input.

[bm1549]

Can we include some unit tests that fail for the old behavior, but pass for the new behavior?

[dgoffredo]

Can we include some unit tests that fail for the old behavior, but pass for the new behavior?

nlohmann::json might fail to encode a string? That's surprising.

Anyway, if the issue was due to an exception thrown by the JSON library, then this might help in testing.

edit: Ah, I bet it's UTF related. Consider auditing the library for other "for convenience" uses of nlohmann::json with string input. In the past I've used it as a shorthand for "give me a quoted, escaped string." If arbitrary byte sequences can trigger an exception, though, that's problematic.

[dmehala]

@dgoffredo Yes, it's a character encoding / serializing issue.

In the past I've used it as a shorthand for "give me a quoted, escaped string."

I was trying to understand what was the purpose, thanks for the context. I will definitely audit for other uses of nlohmann::json.

As always thanks for keeping an eye on dd-trace-cpp David 🙏

[pr-commenter]

Benchmarks

Benchmark execution time: 2024-06-18 08:56:47

Comparing candidate commit d8cd334 in PR branch dmehala/fix-error-handling with baseline commit bb5358e in branch main.

Found 0 performance improvements and 0 performance regressions! Performance is the same for 1 metrics, 0 unstable metrics.