feat(forge): backtraces

[yash-atreya]

Motivation

Closes #2328

Forge lacks backtraces, which are useful for pinpointing the line where the revert/error originated in the call stack. This is especially useful when there are multiple nested calls between different contracts, making traces hard to follow. Backtraces help you easily "goto" the line throwing the error.

Solution

• Introduce backtrace support in forge test
• Backtraces are enabled when verbosity >= 3 and printed if there is a revert
• When verbosity >= 3, we collect the source maps, deployed bytecode, and source content using the BacktraceBuilder.
• In case a test fails, extract the backtraces using from the trace arena using BacktraceBuilder::from_trace(arena); thebytecode.pc (program counter) is mapped to the instruction pointer and then source maps are used to pinpoint the source locations in the call stack.
• Each call is stored as a BacktraceFrame containing the source location of the call along with the contract and function names.

Implementation can be tested using:

• foundryup --pr 11547

• Clone: https://github.com/yash-atreya/test-backtraces

• Run forge t -vvv

• Should work across fresh and cached artifacts, i.e, when there are compiler source ID conflicts - 028b1aa + 5c93b5c

• Fork test backtraces with and without ETHERSCAN_API_KEY c9ce5bc

• Resolve external libs ac1208c

• Handle breaking tests f929071

• Cleanup and remove debug logs

• Handling assertions feat: extend (error) traces to include source code line number and filename #2328 (comment)
  In this case, we are just showing the assertion frame without source location e.g

  foundry/crates/forge/tests/cli/test_cmd.rs

  Line 2990 in d75a675

  at VM.assertEq

Followup

• Internal calls and libraries using solar after feat(cheats): preserve struct order when parsing JSON objects #11612. See:

  foundry/crates/forge/src/cmd/test/mod.rs

  Lines 310 to 327 in 9d4f7eb

  	// Initialize and configure the solar compiler.
  	let mut analysis = solar::sema::Compiler::new(
  	solar::interface::Session::builder().with_stderr_emitter().build(),
  	);
  	let dcx = analysis.dcx_mut();
  	dcx.set_emitter(Box::new(
  	solar::interface::diagnostics::HumanEmitter::stderr(Default::default())
  	.source_map(Some(dcx.source_map().unwrap().clone())),
  	));
  	dcx.set_flags_mut(|f| f.track_diagnostics = false);
  	// Populate solar's global context by parsing and lowering the sources.
  	analysis.enter_mut(|compiler| -> Result<()> {
  	let mut pcx = compiler.parse();
  	configure_pcx(&mut pcx, &config, Some(&project), Some(&analysis_files))?;
  	pcx.parse();
  	let _ = compiler.lower_asts();
  	Ok(())
  	})?;

PR Checklist

• Added Tests
• Added Documentation
• Breaking changes

[DaniPopes]

LGTM, nits.

This does enable steps tracing on -vvv which is 2-3x slower even when no tests fail. I think this is probably fine, however we might want to look into re-executing failing tests since this level of tracing is only needed when they fail.

Edit: we can further optimize this by not requesting stack snapshots since we don't need them. I think we just want jump steps.