cli/tools/coverage/mod.rs - main - githubmirror/deno

cli/tools/coverage/mod.rs
1 046 строк34 KB
Ian Hickson
fix(coverage): count a branch-junction line as covered when either arm runs (#35858)
08 июл 2026, 11:53
08 июл 2026, 11:5395407f0
// Copyright 2018-2026 the Deno authors. MIT license.

use std::fs;
use std::fs::File;
use std::path::Path;
use std::path::PathBuf;
use std::sync::Arc;

use deno_ast::MediaType;
use deno_ast::ModuleKind;
use deno_ast::ModuleSpecifier;
use deno_config::glob::FileCollector;
use deno_config::glob::FilePatterns;
use deno_config::glob::PathOrPattern;
use deno_config::glob::PathOrPatternSet;
use deno_core::anyhow::Context;
use deno_core::anyhow::anyhow;
use deno_core::error::AnyError;
use deno_core::serde_json;
use deno_core::sourcemap::SourceMap;
use deno_core::url::Url;
use deno_resolver::npm::DenoInNpmPackageChecker;
use node_resolver::InNpmPackageChecker;
use regex::Regex;
use reporter::CoverageReporter;
use text_lines::TextLines;

use self::ignore_directives::CoverageComment;
use self::ignore_directives::has_file_ignore_directive;
use self::ignore_directives::lex_comments;
use self::ignore_directives::parse_next_ignore_directives;
use self::ignore_directives::parse_range_ignore_directives;
use crate::args::CliOptions;
use crate::args::CoverageThresholds;
use crate::args::FileFlags;
use crate::args::Flags;
use crate::cdp;
use crate::factory::CliFactory;
use crate::file_fetcher::TextDecodedFile;
use crate::sys::CliSys;
use crate::tools::test::is_supported_test_path;
use crate::util::text_encoding::source_map_from_code;

mod ignore_directives;
mod merge;
mod range_tree;
pub mod reporter;
mod util;
use merge::ProcessCoverage;

#[derive(Debug, Clone)]
struct BranchCoverageItem {
  line_index: usize,
  block_number: usize,
  branch_number: usize,
  taken: Option<i64>,
  is_hit: bool,
}

#[derive(Debug, Clone)]
struct FunctionCoverageItem {
  name: String,
  line_index: usize,
  execution_count: i64,
}

#[derive(Debug, Clone)]
pub struct CoverageReport {
  url: ModuleSpecifier,
  named_functions: Vec<FunctionCoverageItem>,
  branches: Vec<BranchCoverageItem>,
  /// (line_index, number_of_hits)
  found_lines: Vec<(usize, i64)>,
  output: Option<PathBuf>,
}

struct GenerateCoverageReportOptions<'a> {
  script_module_specifier: Url,
  script_media_type: MediaType,
  script_coverage: &'a cdp::ScriptCoverage,
  script_original_source: String,
  script_runtime_source: String,
  maybe_source_map: &'a Option<Vec<u8>>,
  output: &'a Option<PathBuf>,
}

/// The byte offset of the end of the code on the line
/// `line_start_byte_offset..line_end_byte_offset`, with a trailing comment and
/// the whitespace before it removed. A comment does not execute, so it must not
/// decide whether a zero-count range reaches the line's end edge when line hit
/// counts are projected from V8's block ranges. Without this, the same statement
/// is credited to its un-taken branch (0 hits) when nothing follows it and to
/// the covered trailing comment (1 hit) when one does. A line with no trailing
/// comment keeps its raw end, so its attribution is unchanged.
fn line_content_end_byte_offset(
  source: &str,
  line_start_byte_offset: usize,
  line_end_byte_offset: usize,
  comments: &[CoverageComment],
) -> usize {
  let mut end = line_end_byte_offset;
  let mut stripped_comment = false;
  loop {
    let trimmed_end = line_start_byte_offset
      + source[line_start_byte_offset..end].trim_end().len();
    let trailing_comment = comments.iter().find(|comment| {
      comment.range.start >= line_start_byte_offset
        && comment.range.start < trimmed_end
        && comment.range.end >= trimmed_end
    });
    match trailing_comment {
      Some(comment) => {
        stripped_comment = true;
        end = comment.range.start;
      }
      None => {
        if stripped_comment {
          end = trimmed_end;
        }
        return end;
      }
    }
  }
}

fn generate_coverage_report(
  options: GenerateCoverageReportOptions,
) -> Result<CoverageReport, AnyError> {
  let original_comments =
    lex_comments(&options.script_original_source, options.script_media_type);
  let url = Url::parse(&options.script_coverage.url).unwrap();

  if has_file_ignore_directive(&original_comments) {
    return Ok(CoverageReport {
      url,
      named_functions: Vec::new(),
      branches: Vec::new(),
      found_lines: Vec::new(),
      output: options.output.clone(),
    });
  }

  let maybe_source_map = options
    .maybe_source_map
    .as_ref()
    .map(|source_map| SourceMap::from_slice(source_map).unwrap());
  let mut coverage_report = CoverageReport {
    url,
    named_functions: Vec::with_capacity(
      options
        .script_coverage
        .functions
        .iter()
        .filter(|f| !f.function_name.is_empty())
        .count(),
    ),
    branches: Vec::new(),
    found_lines: Vec::new(),
    output: options.output.clone(),
  };

  let original_text_lines = TextLines::new(&options.script_original_source);
  let coverage_ignore_next_directives = parse_next_ignore_directives(
    &original_comments.comments,
    &original_text_lines,
  );
  let coverage_ignore_range_directives = parse_range_ignore_directives(
    &options.script_module_specifier,
    &original_comments.comments,
    &original_text_lines,
  );

  let runtime_comments =
    lex_comments(&options.script_runtime_source, MediaType::JavaScript);
  let runtime_text_lines = TextLines::new(&options.script_runtime_source);
  for function in &options.script_coverage.functions {
    if function.function_name.is_empty() {
      continue;
    }

    // Skip functions that the source map can't trace back to the original
    // source — these are injected by the transformer (e.g. SWC's decorator
    // runtime helpers) and shouldn't show up in user-visible coverage.
    let Some(line_index) = range_to_src_line_index(
      &function.ranges[0],
      &runtime_text_lines,
      &maybe_source_map,
    ) else {
      continue;
    };

    if line_index > 0
      && coverage_ignore_next_directives.contains(&(line_index - 1_usize))
    {
      continue;
    }

    if coverage_ignore_range_directives.iter().any(|range| {
      range.start_line_index <= line_index
        && range.stop_line_index >= line_index
    }) {
      continue;
    }

    coverage_report.named_functions.push(FunctionCoverageItem {
      name: function.function_name.clone(),
      line_index,
      execution_count: function.ranges[0].count,
    });
  }

  for (block_number, function) in
    options.script_coverage.functions.iter().enumerate()
  {
    let block_hits = function.ranges[0].count;

    // For each sub-range, find the parent count: the count of the smallest
    // enclosing range. V8 ranges are nested (ranges[0] is the function body,
    // inner ranges are blocks/branches/loops). The parent count is needed to
    // correctly compute complement branches — e.g. for an if/else inside a
    // loop, the parent is the loop body, not the function.
    let mut parent_counts: Vec<i64> =
      Vec::with_capacity(function.ranges.len().saturating_sub(1));
    for range in &function.ranges[1..] {
      let mut parent_count = block_hits;
      let mut parent_size = usize::MAX;
      for candidate in &function.ranges {
        if std::ptr::eq(candidate, range) {
          continue;
        }
        if candidate.start_char_offset <= range.start_char_offset
          && candidate.end_char_offset >= range.end_char_offset
        {
          let size = candidate.end_char_offset - candidate.start_char_offset;
          if size < parent_size {
            parent_size = size;
            parent_count = candidate.count;
          }
        }
      }
      parent_counts.push(parent_count);
    }

    // Group sub-ranges by their source line to detect branch points.
    // When multiple ranges map to the same source line, they represent
    // different arms of the same branch (e.g. if/else).
    let mut branches_by_line: std::collections::BTreeMap<
      usize,
      Vec<(usize, &cdp::CoverageRange)>,
    > = std::collections::BTreeMap::new();
    for (idx, range) in function.ranges[1..].iter().enumerate() {
      // Skip whitespace-only "gap" ranges V8 emits between blocks (e.g. the
      // sliver between a catch's `return` and `finally`). They aren't real
      // branch arms, so counting them yields phantom branches (#35765).
      let start_byte_offset =
        runtime_text_lines.byte_index_from_char_index(range.start_char_offset);
      let end_byte_offset =
        runtime_text_lines.byte_index_from_char_index(range.end_char_offset);
      if options.script_runtime_source[start_byte_offset..end_byte_offset]
        .trim()
        .is_empty()
      {
        continue;
      }

      // Same rationale as above: drop sub-ranges with no source mapping —
      // they belong to transformer-injected helper code, not the user's
      // source.
      let Some(line_index) =
        range_to_src_line_index(range, &runtime_text_lines, &maybe_source_map)
      else {
        continue;
      };
      branches_by_line
        .entry(line_index)
        .or_default()
        .push((idx, range));
    }

    for (line_index, ranges) in &branches_by_line {
      if *line_index > 0
        && coverage_ignore_next_directives.contains(&(line_index - 1_usize))
      {
        continue;
      }

      if coverage_ignore_range_directives.iter().any(|range| {
        range.start_line_index <= *line_index
          && range.stop_line_index >= *line_index
      }) {
        continue;
      }

      if ranges.len() == 1 {
        let (idx, range) = &ranges[0];
        let enclosing_count = parent_counts[*idx];

        if range.count > enclosing_count {
          // Range executes more often than its enclosing scope — this is a
          // loop body, not a branch arm. Report it without a complement.
          coverage_report.branches.push(BranchCoverageItem {
            line_index: *line_index,
            block_number,
            branch_number: 0,
            taken: if enclosing_count > 0 {
              Some(range.count)
            } else {
              None
            },
            is_hit: range.count > 0,
          });
        } else {
          // Single range at this line: one arm of a branch. The complement
          // (e.g. the else for an if) is implicit with count equal to the
          // enclosing range's count minus this range's count. Using the
          // enclosing (parent) range rather than the function-level count
          // gives correct complements for branches inside loops.
          let taken = if enclosing_count > 0 {
            Some(range.count)
          } else {
            None
          };
          let complement_taken = if enclosing_count > 0 {
            Some(enclosing_count - range.count)
          } else {
            None
          };
          coverage_report.branches.push(BranchCoverageItem {
            line_index: *line_index,
            block_number,
            branch_number: 0,
            taken,
            is_hit: range.count > 0,
          });
          coverage_report.branches.push(BranchCoverageItem {
            line_index: *line_index,
            block_number,
            branch_number: 1,
            taken: complement_taken,
            is_hit: complement_taken.is_some_and(|c| c > 0),
          });
        }
      } else {
        // Multiple ranges at the same line: these are explicit branch arms
        // (e.g. both if and else blocks reported by V8).
        for (branch_number, (_, range)) in ranges.iter().enumerate() {
          let taken = if block_hits > 0 {
            Some(range.count)
          } else {
            None
          };
          coverage_report.branches.push(BranchCoverageItem {
            line_index: *line_index,
            block_number,
            branch_number,
            taken,
            is_hit: range.count > 0,
          });
        }
      }
    }
  }

  // TODO(caspervonb): collect uncovered ranges on the lines so that we can highlight specific
  // parts of a line in color (word diff style) instead of the entire line.
  let mut line_counts = Vec::with_capacity(runtime_text_lines.lines_count());
  for line_index in 0..runtime_text_lines.lines_count() {
    let (line_start_byte_offset, line_end_byte_offset) =
      runtime_text_lines.line_range(line_index);
    let line_start_char_offset =
      runtime_text_lines.char_index(line_start_byte_offset);
    let line_end_char_offset =
      runtime_text_lines.char_index(line_end_byte_offset);
    // The line's end edge for the reset pass below is the end of its code,
    // excluding a trailing comment so a comment cannot change what the line is
    // credited with.
    let content_end_byte_offset = line_content_end_byte_offset(
      &options.script_runtime_source,
      line_start_byte_offset,
      line_end_byte_offset,
      &runtime_comments.comments,
    );
    let line_content_end_char_offset =
      runtime_text_lines.char_index(content_end_byte_offset);
    let ignore = runtime_comments.comments.iter().any(|comment| {
      comment.range.start <= line_start_byte_offset
        && comment.range.end >= line_end_byte_offset
    }) || options.script_runtime_source
      [line_start_byte_offset..line_end_byte_offset]
      .trim()
      .is_empty();
    let mut count = 0;

    if ignore {
      count = 1;
    } else {
      // Find the count from the most specific (smallest/innermost) range that
      // fully covers this line. V8 coverage ranges are nested: ranges[0] is the
      // function body and ranges[1..] are inner blocks/branches. The innermost
      // range that covers a line gives the correct execution count for that line.
      let mut best_range_size = usize::MAX;
      for function in &options.script_coverage.functions {
        for range in &function.ranges {
          if range.start_char_offset <= line_start_char_offset
            && range.end_char_offset >= line_end_char_offset
          {
            let range_size = range.end_char_offset - range.start_char_offset;
            if range_size < best_range_size {
              best_range_size = range_size;
              count = range.count;
            }
          }
        }
      }

      // Reset the count when a zero-count range genuinely covers this line's
      // code, judged three ways below. Merely reaching one edge is not enough:
      // where a covered arm and a zero-count sibling arm meet on one line, the
      // covered side is real code that ran and must not be hidden. That is the
      // case at a branch junction such as `} else {`, `} catch {`, or
      // `} finally {`: when only one arm ran, the other arm's zero-count range
      // clips the opposite edge, and zeroing the whole line on that single edge
      // is what otherwise makes the junction count as covered only when both
      // arms run in the same coverage process. Splitting the two arms across
      // separate test files then drops the junction from the merged line
      // coverage. A zero-count range floating in the middle of a line (matching
      // none of the tests) is a V8 block-boundary gap, e.g. the unreachable path
      // between a catch's return and finally, and is ignored.
      for (range_index, range) in options
        .script_coverage
        .functions
        .iter()
        .flat_map(|function| function.ranges.iter().enumerate())
      {
        if range.count > 0 {
          continue;
        }

        let overlaps = range.start_char_offset < line_end_char_offset
          && range.end_char_offset > line_start_char_offset;
        if !overlaps {
          continue;
        }

        // The line sits entirely inside the uncovered range: a fully uncovered
        // line, or one in the middle of a multi-line uncovered block.
        let spans_content = range.start_char_offset <= line_start_char_offset
          && range.end_char_offset >= line_content_end_char_offset;
        // An uncovered statement confined to this line that runs to the end of
        // its code, e.g. the never-taken `throw` in `if (!x) throw …`. A
        // multi-line block whose brace only clips a junction extends past the
        // line end and is excluded here, so the covered half survives.
        let inline_uncovered = range.start_char_offset
          >= line_start_char_offset
          && range.end_char_offset <= line_end_char_offset
          && range.end_char_offset >= line_content_end_char_offset;
        // A never-called function is a single count-0 range (ranges[0]) covering
        // the whole function. Its signature line stays uncovered like its body,
        // even though the range starts a few characters into that line, after an
        // `export`/`async` keyword, which the span test above would miss.
        let whole_function = range_index == 0
          && (range.start_char_offset <= line_start_char_offset
            || range.end_char_offset >= line_content_end_char_offset);
        if spans_content || inline_uncovered || whole_function {
          count = 0;
        }
      }
    }

    line_counts.push(count);
  }

  let found_lines_coverage_filter = |(line, _): &(usize, i64)| -> bool {
    if coverage_ignore_range_directives.iter().any(|range| {
      range.start_line_index <= *line && range.stop_line_index >= *line
    }) {
      return false;
    }

    if coverage_ignore_next_directives.contains(line) {
      return false;
    }

    if *line == 0_usize {
      return true;
    }

    if coverage_ignore_next_directives.contains(&(line - 1_usize)) {
      return false;
    }

    true
  };

  coverage_report.found_lines =
    if let Some(source_map) = maybe_source_map.as_ref() {
      let script_runtime_source_lines =
        options.script_runtime_source.lines().collect::<Vec<_>>();
      let mut found_lines = line_counts
        .iter()
        .enumerate()
        .flat_map(|(index, count)| {
          // get all the mappings from this destination line to a different src line
          let mut results = source_map
            .tokens()
            .filter(|token| {
              let dst_line = token.get_dst_line() as usize;
              dst_line == index && {
                let dst_col = token.get_dst_col() as usize;
                let content = script_runtime_source_lines
                  .get(dst_line)
                  .and_then(|line| {
                    line.get(dst_col..std::cmp::min(dst_col + 2, line.len()))
                  })
                  .unwrap_or("");

                !content.is_empty()
                  && content != "/*"
                  && content != "*/"
                  && content != "//"
              }
            })
            .map(move |token| (token.get_src_line() as usize, *count))
            .collect::<Vec<_>>();
          // only keep the results that point at different src lines
          results.sort_unstable_by_key(|(index, _)| *index);
          results.dedup_by_key(|(index, _)| *index);
          results.into_iter()
        })
        .filter(found_lines_coverage_filter)
        .collect::<Vec<(usize, i64)>>();

      found_lines.sort_unstable_by_key(|(index, _)| *index);
      // combine duplicated lines - when multiple compiled JS lines map to the
      // same source line, use the maximum count rather than summing, since
      // hitting the same source line from different compiled lines doesn't mean
      // it was executed more times.
      for i in (1..found_lines.len()).rev() {
        if found_lines[i].0 == found_lines[i - 1].0 {
          found_lines[i - 1].1 =
            std::cmp::max(found_lines[i - 1].1, found_lines[i].1);
          found_lines.remove(i);
        }
      }
      found_lines
    } else {
      line_counts
        .into_iter()
        .enumerate()
        .filter(found_lines_coverage_filter)
        .collect::<Vec<(usize, i64)>>()
    };

  Ok(coverage_report)
}

/// Maps a runtime coverage range back to a line in the original source.
///
/// Returns `None` when the source map exists but has no mapping for this
/// position. That happens for code injected by the transformer (e.g. SWC's
/// decorator runtime helpers like `applyDecs2203R`) — the bytes exist in
/// the runtime JS but didn't come from the user's source, so callers
/// should drop these coverage entries rather than reporting them as if
/// they were the user's own code.
fn range_to_src_line_index(
  range: &cdp::CoverageRange,
  text_lines: &TextLines,
  maybe_source_map: &Option<SourceMap>,
) -> Option<usize> {
  let source_lc = text_lines.line_and_column_index(
    text_lines.byte_index_from_char_index(range.start_char_offset),
  );
  if let Some(source_map) = maybe_source_map.as_ref() {
    source_map
      .lookup_token(source_lc.line_index as u32, source_lc.column_index as u32)
      .map(|token| token.get_src_line() as usize)
  } else {
    Some(source_lc.line_index)
  }
}

fn collect_coverages(
  cli_options: &CliOptions,
  files: FileFlags,
  initial_cwd: &Path,
) -> Result<Vec<cdp::ScriptCoverage>, AnyError> {
  let mut coverages: Vec<cdp::ScriptCoverage> = Vec::new();
  let file_patterns = FilePatterns {
    base: initial_cwd.to_path_buf(),
    include: Some({
      if files.include.is_empty() {
        PathOrPatternSet::new(vec![PathOrPattern::Path(
          initial_cwd.to_path_buf(),
        )])
      } else {
        PathOrPatternSet::from_include_relative_path_or_patterns(
          initial_cwd,
          &files.include,
        )?
      }
    }),
    exclude: PathOrPatternSet::new(vec![]),
  };
  let file_paths = FileCollector::new(|e| {
    e.path.extension().map(|ext| ext == "json").unwrap_or(false)
  })
  .ignore_git_folder()
  .ignore_node_modules()
  .set_vendor_folder(cli_options.vendor_dir_path().map(ToOwned::to_owned))
  .collect_file_patterns(&CliSys::default(), &file_patterns);

  let coverage_patterns = FilePatterns {
    base: initial_cwd.to_path_buf(),
    include: None,
    exclude: PathOrPatternSet::from_exclude_relative_path_or_patterns(
      initial_cwd,
      &files.ignore,
    )
    .context("Invalid ignore pattern.")?,
  };

  for file_path in file_paths {
    let new_coverage = fs::read_to_string(file_path.as_path())
      .map_err(AnyError::from)
      .and_then(|json| {
        serde_json::from_str::<cdp::ScriptCoverage>(&json)
          .map_err(AnyError::from)
      })
      .with_context(|| format!("Failed reading '{}'", file_path.display()))?;
    let url = Url::parse(&new_coverage.url)?;
    if coverage_patterns.matches_specifier(&url) {
      coverages.push(new_coverage);
    }
  }

  coverages.sort_by_key(|k| k.url.clone());

  Ok(coverages)
}

fn filter_coverages(
  coverages: Vec<cdp::ScriptCoverage>,
  include: Vec<String>,
  exclude: Vec<String>,
  in_npm_pkg_checker: &DenoInNpmPackageChecker,
  link_dir_urls: &[String],
) -> Vec<cdp::ScriptCoverage> {
  let include: Vec<Regex> =
    include.iter().map(|e| Regex::new(e).unwrap()).collect();

  let exclude: Vec<Regex> =
    exclude.iter().map(|e| Regex::new(e).unwrap()).collect();

  // Matches virtual file paths for doc testing
  // e.g. file:///path/to/mod.ts#23-29.ts
  let doc_test_re =
    Regex::new(r"#\d+-\d+\.(js|mjs|cjs|jsx|ts|mts|cts|tsx)$").unwrap();

  coverages
    .into_iter()
    .filter(|e| {
      let is_internal = e.url.starts_with("ext:")
        || e.url.starts_with("data:")
        || e.url.starts_with("blob:")
        || e.url.ends_with("__anonymous__")
        || e.url.ends_with("$deno$test.mjs")
        || e.url.contains("/$deno$stdin.")
        || e.url.ends_with(".snap")
        || is_supported_test_path(Path::new(e.url.as_str()))
        || doc_test_re.is_match(e.url.as_str())
        // Exclude packages brought in via the "links" (formerly "patch")
        // feature. These are third-party dependencies replacing a registry
        // version, so they shouldn't show up in the user's coverage report,
        // matching how npm packages are excluded above.
        || link_dir_urls.iter().any(|dir| e.url.starts_with(dir))
        || Url::parse(&e.url)
          .ok()
          .map(|url| in_npm_pkg_checker.in_npm_package(&url))
          .unwrap_or(false);

      let is_included = include.iter().any(|p| p.is_match(&e.url));
      let is_excluded = exclude.iter().any(|p| p.is_match(&e.url));

      (include.is_empty() || is_included) && !is_excluded && !is_internal
    })
    .collect::<Vec<cdp::ScriptCoverage>>()
}

#[allow(clippy::too_many_arguments, reason = "coverage entry point")]
pub fn cover_files(
  flags: Arc<Flags>,
  files_include: Vec<String>,
  files_ignore: Vec<String>,
  include: Vec<String>,
  exclude: Vec<String>,
  output: Option<String>,
  cli_threshold: Option<f64>,
  reporters: &[&dyn CoverageReporter],
) -> Result<(), AnyError> {
  if files_include.is_empty() {
    return Err(anyhow!("No matching coverage profiles found"));
  }

  let factory = CliFactory::from_flags(flags);
  let cli_options = factory.cli_options()?;
  let in_npm_pkg_checker = factory.in_npm_pkg_checker()?;
  let file_fetcher = factory.file_fetcher()?;
  let emitter = factory.emitter()?;
  let cjs_tracker = factory.cjs_tracker()?;

  // Use the first include path as the default output path.
  let coverage_root = cli_options.initial_cwd().join(&files_include[0]);
  let script_coverages = collect_coverages(
    cli_options,
    FileFlags {
      include: files_include,
      ignore: files_ignore,
    },
    cli_options.initial_cwd(),
  )?;
  if script_coverages.is_empty() {
    return Err(anyhow!("No coverage files found"));
  }
  let link_dir_urls = cli_options
    .workspace()
    .link_folders()
    .keys()
    .map(|url| url.as_str().to_string())
    .collect::<Vec<_>>();
  let script_coverages = filter_coverages(
    script_coverages,
    include,
    exclude,
    in_npm_pkg_checker,
    &link_dir_urls,
  );
  if script_coverages.is_empty() {
    return Err(anyhow!("No covered files included in the report"));
  }

  let proc_coverages: Vec<_> = script_coverages
    .into_iter()
    .map(|mut cov| {
      // Merge the same module loaded under different specifiers. Fragments
      // never affect the source, so strip them for any file/http(s) URL.
      // Query strings are only stripped for `file:` (same file on disk);
      // http(s) servers can return different bytes per query and the cache
      // keys on the full URL, so keep it there.
      if let Ok(mut url) = Url::parse(&cov.url) {
        let strip_fragment = matches!(url.scheme(), "file" | "http" | "https")
          && url.fragment().is_some();
        let strip_query = url.scheme() == "file" && url.query().is_some();
        if strip_fragment {
          url.set_fragment(None);
        }
        if strip_query {
          url.set_query(None);
        }
        if strip_fragment || strip_query {
          cov.url = url.to_string();
        }
      }
      ProcessCoverage { result: vec![cov] }
    })
    .collect();

  let script_coverages = if let Some(c) = merge::merge_processes(proc_coverages)
  {
    c.result
  } else {
    vec![]
  };

  let out_mode = match output {
    Some(ref path) => match File::create(path) {
      Ok(_) => Some(PathBuf::from(path)),
      Err(e) => {
        return Err(anyhow!("Failed to create output file: {}", e));
      }
    },
    None => None,
  };
  let get_message = |specifier: &ModuleSpecifier| -> String {
    format!(
      "Source not found for \"{}\" (was it deleted after coverage was collected?). Skipping.",
      specifier,
    )
  };

  let mut file_reports = Vec::with_capacity(script_coverages.len());

  for script_coverage in script_coverages {
    let module_specifier = deno_path_util::resolve_url_or_path(
      &script_coverage.url,
      cli_options.initial_cwd(),
    )?;

    let maybe_file_result =
      file_fetcher.get_cached_source_or_local(&module_specifier);
    let file = match maybe_file_result {
      Ok(Some(file)) => TextDecodedFile::decode(file)?,
      Ok(None) => {
        log::warn!("{}", get_message(&module_specifier),);
        continue;
      }
      Err(err) => {
        log::warn!("{}: {}", get_message(&module_specifier), err,);
        continue;
      }
    };

    let original_source = file.source.clone();
    // Check if file was transpiled
    let transpiled_code = match file.media_type {
      MediaType::JavaScript
      | MediaType::Unknown
      | MediaType::Css
      | MediaType::Html
      | MediaType::Markdown
      | MediaType::Sql
      | MediaType::Wasm
      | MediaType::Cjs
      | MediaType::Mjs
      | MediaType::Json
      | MediaType::Jsonc
      | MediaType::Json5 => None,
      MediaType::Dts | MediaType::Dmts | MediaType::Dcts => Some(String::new()),
      MediaType::TypeScript
      | MediaType::Jsx
      | MediaType::Mts
      | MediaType::Cts
      | MediaType::Tsx => {
        let module_kind = ModuleKind::from_is_cjs(
          cjs_tracker.is_maybe_cjs(&file.specifier, file.media_type)?,
        );
        Some(
          match emitter.maybe_cached_emit(
            &file.specifier,
            module_kind,
            &file.source,
          )? {
            Some(code) => code,
            None => {
              log::warn!(
                "Missing transpiled source code for: \"{}\" (was it deleted after coverage was collected?). Skipping.",
                file.specifier,
              );
              continue;
            }
          },
        )
      }
      MediaType::SourceMap => {
        unreachable!()
      }
    };
    let runtime_code: String = match transpiled_code {
      Some(code) => code,
      None => original_source.to_string(),
    };

    let source_map = source_map_from_code(runtime_code.as_bytes());
    let coverage_report =
      generate_coverage_report(GenerateCoverageReportOptions {
        script_module_specifier: module_specifier.clone(),
        script_media_type: file.media_type,
        script_coverage: &script_coverage,
        script_original_source: original_source.to_string(),
        script_runtime_source: runtime_code.as_str().to_owned(),
        maybe_source_map: &source_map,
        output: &out_mode,
      })
      .with_context(|| {
        format!(
          "Failed to generate coverage report for file ({module_specifier})"
        )
      })?;

    if !coverage_report.found_lines.is_empty() {
      file_reports.push((coverage_report, original_source.to_string()));
    }
  }

  // All covered files, might have had ignore directive and we can end up
  // with no reports at this point.
  if file_reports.is_empty() {
    return Err(anyhow!("No covered files included in the report"));
  }

  for reporter in reporters {
    reporter.done(&coverage_root, &file_reports);
  }

  // Layer the `--threshold` CLI flag over per-metric thresholds from deno.json
  // (the flag wins) and fail the command if any configured threshold is unmet.
  let config_thresholds = cli_options.resolve_coverage_thresholds()?;
  let thresholds = CoverageThresholds {
    lines: cli_threshold.or(config_thresholds.lines),
    branches: cli_threshold.or(config_thresholds.branches),
    functions: cli_threshold.or(config_thresholds.functions),
  };
  check_coverage_thresholds(&file_reports, thresholds)?;

  Ok(())
}

/// Computes the aggregate line, branch, and function coverage percentages
/// across all reported files. A metric with no measurable items counts as 100%
/// (e.g. a `branches` threshold passes vacuously for files that have no
/// branches). Reuses the same accumulation and percentage helpers as the
/// summary reporter so the checked numbers match the printed ones.
fn aggregate_coverage_percentages(
  file_reports: &[(CoverageReport, String)],
) -> (f64, f64, f64) {
  let mut stats = reporter::CoverageStats::default();
  for (report, _) in file_reports {
    stats.add_report(report);
  }
  let percent = |hit: usize, miss: usize| -> f64 {
    util::calc_coverage_display_info(hit, miss).1 as f64
  };
  (
    percent(stats.line_hit, stats.line_miss),
    percent(stats.branch_hit, stats.branch_miss),
    percent(stats.fn_hit, stats.fn_miss),
  )
}

fn check_coverage_thresholds(
  file_reports: &[(CoverageReport, String)],
  thresholds: CoverageThresholds,
) -> Result<(), AnyError> {
  if thresholds.lines.is_none()
    && thresholds.branches.is_none()
    && thresholds.functions.is_none()
  {
    return Ok(());
  }

  let (line_percent, branch_percent, fn_percent) =
    aggregate_coverage_percentages(file_reports);

  let mut failures = Vec::new();
  let mut check = |name: &str, actual: f64, threshold: Option<f64>| {
    if let Some(threshold) = threshold
      && actual < threshold
    {
      // Floor the displayed value to two decimals so a near-miss like 89.999%
      // doesn't render as "90.00% is below the threshold of 90.00%".
      let actual = (actual * 100.0).floor() / 100.0;
      failures.push(format!(
        "  - {name} coverage {actual:.2}% is below the threshold of {threshold:.2}%"
      ));
    }
  };
  check("Line", line_percent, thresholds.lines);
  check("Branch", branch_percent, thresholds.branches);
  check("Function", fn_percent, thresholds.functions);

  if failures.is_empty() {
    Ok(())
  } else {
    Err(
      CoverageThresholdError(format!(
        "Coverage threshold not met:\n{}",
        failures.join("\n")
      ))
      .into(),
    )
  }
}

/// Returned when a configured coverage threshold is not met. A distinct type so
/// `deno test --coverage` can treat it as fatal while still tolerating benign
/// report-generation errors.
#[derive(Debug)]
pub struct CoverageThresholdError(pub String);

impl std::fmt::Display for CoverageThresholdError {
  fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
    write!(f, "{}", self.0)
  }
}

impl std::error::Error for CoverageThresholdError {}

#[cfg(test)]
mod tests {
  use super::*;

  /// The source substring credited as code on `line_index`: the line with any
  /// trailing comment and the whitespace before it removed from its end.
  /// Exercises `line_content_end_byte_offset` with real lexer comment ranges.
  fn code_on_line(source: &str, line_index: usize) -> &str {
    let comments = lex_comments(source, MediaType::JavaScript).comments;
    let text_lines = TextLines::new(source);
    let (start, end) = text_lines.line_range(line_index);
    let content_end =
      line_content_end_byte_offset(source, start, end, &comments);
    &source[start..content_end]
  }

  #[test]
  fn no_trailing_comment_keeps_whole_line() {
    assert_eq!(
      code_on_line("if (!x) throw new Error(\"e\");", 0),
      "if (!x) throw new Error(\"e\");",
    );
  }

  #[test]
  fn strips_trailing_line_comment() {
    assert_eq!(
      code_on_line("if (!x) throw new Error(\"e\"); // never taken", 0),
      "if (!x) throw new Error(\"e\");",
    );
  }

  #[test]
  fn strips_trailing_block_comment() {
    assert_eq!(code_on_line("foo(); /* done */", 0), "foo();");
  }

  #[test]
  fn strips_multiple_stacked_trailing_comments() {
    assert_eq!(code_on_line("foo(); /* a */ // b", 0), "foo();");
  }

  #[test]
  fn keeps_interior_comment_when_code_follows() {
    assert_eq!(
      code_on_line("foo(/* inline */ x);", 0),
      "foo(/* inline */ x);"
    );
  }

  #[test]
  fn keeps_leading_indentation() {
    assert_eq!(code_on_line("  return 1; // done", 0), "  return 1;");
  }

  #[test]
  fn comment_only_line_has_no_code() {
    assert_eq!(code_on_line("  // just a comment", 0), "");
  }
}