Struct TestSession 

pub struct TestSession { /* private fields */ }

One checker for a whole run: every test file gets its own fresh program state (globals, package.loaded, module state), but modules are type-checked and generated once and served to every file from the checker’s store.

Implementations

impl TestSession

pub fn new( lint_spec: Option<&str>, lib: &str, filter: Option<&str>, opts: RunOptions, ) -> Result<Self>

pub fn checker(&self) -> &Htl

The session’s checker (for Htl::executable_ranges on the sources a run touched).

pub fn run_file(&self, path: &Path) -> Result<FileReport>

Run one file in a fresh program state borrowing the session’s checker. The checker’s search path is restored afterwards so files do not see each other’s directories.

pub fn run_file_with( &self, path: &Path, generated: Option<(&str, &CheckInfo)>, preload: &[(String, String, PathBuf)], ) -> Result<(FileReport, Option<String>)>

Run one file, reusing Lua the caller generated earlier.

generated is (the Lua, what checking it reported). Everything before the codegen still happens — the searcher, the search path, the project and the config all have to be in place before the code can execute — and everything after it happens as usual. Only the check and the codegen are skipped.

The run itself is never reused, and this signature cannot express reusing it: a test has to run to say whether it passes.

Returns the report and, when this call generated the Lua rather than being handed it, that Lua — so a caller keeping a cache has something to keep. preload is (module name, its generated Lua, the file it came from) for modules this file will require. Each one goes in front of the searcher, so requiring it does not check and generate it during the run. A module not in the list still loads the usual way; the list is an optimisation, never a restriction on what can be required.