`ion-tests` needs a better way to capture encoding context

[zslayton]

See this comment:

ion-rust/tests/conformance_dsl/fragment.rs

Lines 422 to 459 in 8935e5c

	// The Writer is asked to serialize Elements that represent system values.
	// In many cases, it will emit a macro table containing macros that are not in its own
	// encoding context. For example, this snippet:
	//
	// (mactab (macro m (v!) v))
	//
	// will emit a system value that the reader will understand defines macro `m`.
	// However, the writer emitting the system value does not have macro `m` in its own
	// encoding context. If the test later invokes `m` like this:
	//
	// (toplevel ('#$:m' 5))
	//
	// the Writer will raise an error reporting that the test is trying to invoke a
	// non-existent macro.
	//
	// As a workaround, when serializing top-level Elements, the writer does an initial
	// serialization pass to load any necessary encoding context changes.
	//
	// Serialize the data once...
	let serialized = v1_1::Text::encode_all(self.elems.as_slice())?;
	// ...then read the data, constructing a macro table in the Reader...
	let mut reader = Reader::new(AnyEncoding, serialized)?;
	while reader.next()?.is_some() {}
	let macro_table = reader.macro_table();
	if ctx.version() == IonVersion::V1_1 {
	// For each macro in the Reader...
	for mac in macro_table.iter() {
	// ...try to register the macro in the Writer. For simplicity, we skip over
	// system macros that are already defined.
	let _result = writer.register_macro(&mac);
	}
	}
	// Now that the Writer has the necessary context, it can encode the ProxyElements.
	for elem in self.elems.as_slice() {
	writer.write(ProxyElement(elem, ctx))?;
	}
	Ok(())

cc @nirosys

[nirosys]

I bumped into this when trying to merge #953 (comment). I'm working to address the test failures, which currently looks like it will require intercepting the macro definitions in order to register them and skip re-writing the definition in the fragment.

[nirosys]

Found another issue in ion-tests/conformance/demos/telemetry_log.ion where set_macros is invoked. This is a bit terrifying.. given that set_macros could be anywhere. This also applies to add_macros, add_symbols, and set_symbols.

Looking to see how set_macros is implemented.. and if the writer would handle registering the macros itself should a user write a set_macros invocation and then try to write invocations of those macros.