Better provenance metadata

[agriyakhetarpal]

This PR updates the provenance metadata for all tests. We now record provenance as a structured object that includes the test's node ID, the commit hash, and a direct URL to the relevant source lines in https://github.com/conda/conda/.

AI disclosure

I quickly realised after updating a few test file links manually that this was going to be a tedious task, so the update (see the second commit) was performed by a script that Claude Code, using the Sonnet 4.6 model, wrote upon my asking. Here is the script:

import ast, re, urllib.request, glob as g

COMMIT = "03329e0f4a627c9b9aa92ef34f7f93b9aa83e438"
BASE = f"https://raw.githubusercontent.com/conda/conda/{COMMIT}"

# 1. Collect all node_ids from YAML files
yaml_files = g.glob("conda-solver-tests/*.yaml")
node_ids = set()
for yf in yaml_files:
    with open(yf) as f:
        for line in f:
            m = re.match(r'\s+node_id:\s+(.+)', line)
            if m:
                node_ids.add(m.group(1).strip())

# 2. Group by source file
file_funcs = {}
for nid in node_ids:
    parts = nid.split("::")
    filepath, func_part = parts[0], parts[1]
    file_funcs.setdefault(filepath, set()).add(func_part)

# 3. Fetch files and parse with AST to find function line ranges
func_lines = {}
for filepath, funcs in file_funcs.items():
    url = f"{BASE}/{filepath}"
    with urllib.request.urlopen(url) as resp:
        source = resp.read().decode("utf-8")
    tree = ast.parse(source)
    for node in ast.walk(tree):
        # Class methods (e.g. SolverTests.test_iopro_mkl)
        if isinstance(node, ast.ClassDef):
            for item in node.body:
                if isinstance(item, (ast.FunctionDef, ast.AsyncFunctionDef)):
                    qualified = f"{node.name}.{item.name}"
                    if qualified in funcs:
                        func_lines[(filepath, qualified)] = (item.lineno, item.end_lineno)
        # Top-level functions
        if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef)):
            if node.name in funcs:
                func_lines[(filepath, node.name)] = (node.lineno, node.end_lineno)

# 4. Patch YAML files: update url lines based on preceding node_id
for yf in yaml_files:
    with open(yf) as f:
        lines = f.readlines()
    new_lines = []
    current_node_id = None
    for line in lines:
        m = re.match(r'(\s+)node_id:\s+(.+)', line)
        if m:
            current_node_id = m.group(2).strip()
        url_m = re.match(r'(\s+)url:\s+(https://github\.com/conda/conda/blob/.+)', line)
        if url_m and current_node_id:
            indent = url_m.group(1)
            parts = current_node_id.split("::")
            key = (parts[0], parts[1])
            if key in func_lines:
                start, end = func_lines[key]
                new_url = f"https://github.com/conda/conda/blob/{COMMIT}/{parts[0]}#L{start}-L{end}"
                new_lines.append(f"{indent}url: {new_url}\n")
                current_node_id = None
                continue
        new_lines.append(line)
    with open(yf, 'w') as f:
        f.writelines(new_lines)
    print(f"Updated {yf}")

We could restructure the above script into a custom pre-commit hook manually, so that I don't have to come back to it again and do all of this later. Thoughts, @jaimergp?

[jaimergp]

I kind of like it 😂 You mean a pre-commit hook, right? As long as it's not flaky, yep, why not. We can also add a meta-test specifically for that, as you wish. Maybe that's easier than a pre-commit hook.

[agriyakhetarpal]

I kind of like it 😂 You mean a pre-commit hook, right? As long as it's not flaky, yep, why not. We can also add a meta-test specifically for that, as you wish. Maybe that's easier than a pre-commit hook.

Ah, yes. I do mean a pre-commit hook. I realise I missed writing that in the PR description after the word "custom" 😅 Let me explore if a meta-test is easier/better; maybe it is.

That said, the linters are not running in this PR. Could you please install the pre-commit.ci app to this repository for me?

[agriyakhetarpal]

I spent some time adding a meta-test of sorts that runs a bunch of checks for the provenance data. This was a bit tedious, so agentic coding tools came in a bit handy here. However, I am aware that there is a bit more code than we'd imagine – we could loosen the validations a bit, if you'd like. I kind of like this a bit better than having a pre-commit hook, since it's executed with the tests.