fix: support large files in comparisons

Replace the 16 MiB comparison rejection with independent bounded per-side loading and frozen Working blobs. Bound diff computation, defer large inline highlighting, index source-row navigation, and expose actual load failures. Verified the real 52.3 MB parser through repository/session APIs and terminal rendering with EOF/start navigation.

TODO.md

@@ -1,5 +1,6 @@
 - [x] syntax highlighting for Honey (honeylang)
     - honeylang treesitter project located at `/Users/valdemar/Developer/Personal/Languages/tree-sitter-honey`.
+- [ ] bug: `x` cut action does not copy the hovered character if not in selection mode, though should always copy.
 - [ ] diff view: GitHub code review integration after the local review UX is usable.
 	- publish local reviews explicitly: general PR comments, inline line/range comments, review summaries, approval, and requests for changes.
 	- preserve explicit PR identity and reviewed commit IDs; validate anchors against GitHub's PR diff and recheck the PR head before submission.

docs/development.md

@@ -101,7 +101,8 @@ Diff styling checks cover independent snapshot language/shebang/injection detect
 - PTY throughput alone cannot prove terminal presentation stays smooth: capture the editor and terminal while the user reproduces a pause. Synchronized-output commands must not reach the terminal during frame computation, even when individual frames and queued-input catch-up appear fast.
 - Annotation coordinate validation fast-paths ASCII grapheme lengths using byte lengths with CRLF pair accounting, retaining Unicode segmentation for non-ASCII text. Validation remains atomic and still checks every token; large Unicode buffers or diagnostic batches can require foreground work.
 - Legacy synchronous plugin invocation may pause input for up to 500 ms before a host timeout. Async plugins opt into background replies; pending requests and transport queues are bounded. Config/plugin startup has a 2-second deadline per host. Buffer context/document snapshots must fit within a 4 MiB JSON frame; a too-large frame disables that plugin, not editing.
-- Repository diff currently requires Unix subprocess-group cancellation and Git-backed jj (or Git alone). Each file is limited to 16 MiB; each subprocess output stream and Working/conflict snapshot to 64 MiB; each command to 60 seconds. Limit, transport, and non-commit-tag errors are explicit rather than silently truncated. Temporary comparison stores are removed when the diff closes.
+- Repository diff requires Unix subprocess-group cancellation and Git-backed jj (or Git alone). Selected comparison sides support 64 MiB and 4,000,000 source lines each; frozen Working snapshots support 512 MiB total. Subprocess streams remain bounded to 64 MiB and commands to 60 seconds. Working text is frozen into private Git blobs; catalogs stay lazy and selected blobs load separately without concatenated batch copies. Limits and transport errors are explicit; temporary stores are removed on close. Picker/gutter annotation budgets remain separate.
+- Large line comparisons skip equal prefixes/suffixes before allocating diff tokens; added/removed files bypass the diff algorithm. Myers work has a 100 ms deadline, after which unresolved regions can be conservatively shown as replacements without dropping text. Inline highlighting is viewport-lazy above 20,000 combined source rows; large/pathological lines use whole-line change shading. Source/display navigation uses indexed contiguous ranges rather than scanning all preceding rows.
 - Normal-picker status uses the gutter worker's five-second command and 16 MiB stream limits. Catalogs are bounded to 100,000 paths; dirty-buffer batches to 1,024 files/64 MiB; individual overlays, conflict files, and rename candidates to 16 MiB; conflict/rename data to 64 MiB. Temporary baseline stores are retained for warm metadata polling and removed when the worker stops.
 
 See [syntax caches](architecture.md#syntax-caches) for the current full-tree highlighting ceiling.

docs/editing.md

@@ -517,6 +517,8 @@ Both diff panes apply available semantic tokens from their own buffer, independe
 
 Binary data, missing final newlines, permission/type changes, symlink targets, and submodule pointers have explicit notices. Symlinks are not followed and submodule contents are not traversed. jj conflicts use jj's file-content projection and carry conflict notices; ordinary files remain readable. Identical-content moves are shown as renames; modified moves appear as an added and removed file.
 
+Large comparisons load each selected side independently, up to 64 MiB and 4,000,000 source lines per side. Frozen Working snapshots have a 512 MiB aggregate budget. Exceeding a limit displays the actual error in the document pane. Large files use the normal plain-text syntax fallback; pathological comparisons may show broader replacement regions or whole-line highlights rather than block input searching for a minimal diff. Full source text remains available.
+
 ## Jujutsu conflict resolution
 
 Run **`:resolve`** in Normal mode for the current jj working copy. It opens a dedicated tab and selects the current conflicted file (otherwise the first conflict). Historical comparisons remain read-only; resolving never implicitly checks out or creates another change.

src/diff.rs

@@ -1,8 +1,10 @@
 use crate::editor::{BackspaceKind, Input};
 use crate::files::{FilePicker, FilePresentation, FileView, fuzzy_match};
 use std::{
+    borrow::Cow,
     collections::{BTreeMap, BTreeSet},
     path::{Path, PathBuf},
+    time::{Duration, Instant},
 };
 use unicode_segmentation::UnicodeSegmentation;
 use unicode_width::UnicodeWidthStr;
@@ -281,11 +283,11 @@ pub(crate) struct ColumnLine<'a> {
     pub changed: bool,
     pub emphasize: bool,
     pub newline: bool,
-    changes: &'a [[std::ops::Range<usize>; 2]],
+    changes: Cow<'a, [[std::ops::Range<usize>; 2]]>,
 }
 
 impl<'a> ColumnLine<'a> {
-    pub fn changes(&self) -> impl Iterator<Item = &'a std::ops::Range<usize>> + 'a {
+    pub fn changes(&self) -> impl Iterator<Item = &std::ops::Range<usize>> + '_ {
         let side = usize::from(self.side == Focus::After);
         let emphasize = self.emphasize;
         self.changes
@@ -363,7 +365,21 @@ impl<'a> ColumnView<'a> {
                 || (!before && row.kind == RowKind::Added),
             emphasize: !matches!(self.file.status, FileStatus::Added | FileStatus::Removed),
             newline: source_row + 1 < source.len() || eof,
-            changes: &row.changes,
+            changes: if self.diff.lazy_changes && row.kind == RowKind::Changed {
+                match (row.before, row.after) {
+                    (Some(_), Some(_)) if self.file.hunks.is_empty() => {
+                        // A stitched History base is changed by another column.
+                        Cow::Owned(vec![[0..usize::MAX, 0..usize::MAX]])
+                    }
+                    (Some(before), Some(after)) => Cow::Owned(character_changes(
+                        &self.file.before[before],
+                        &self.file.after[after],
+                    )),
+                    _ => Cow::Borrowed(&row.changes),
+                }
+            } else {
+                Cow::Borrowed(&row.changes)
+            },
         }))
     }
 
@@ -480,6 +496,7 @@ impl HistoryComparison {
         }
         for column in &mut columns {
             column.display_rows = (0..column.rows.len()).map(DisplayRow::Line).collect();
+            column.rebuild_source_ranges();
         }
         Self { columns }
     }
@@ -561,6 +578,8 @@ pub struct DiffState {
     pub rows: Vec<AlignedRow>,
     pub display_rows: Vec<DisplayRow>,
     display_sides: Vec<Option<Focus>>,
+    source_ranges: [Vec<(std::ops::Range<usize>, usize)>; 2],
+    lazy_changes: bool,
     /// Native empty terminal rows explicitly supplied by resolver text.
     terminal_rows: [Option<usize>; 2],
     excerpt_ranges: Vec<(usize, usize)>,
@@ -621,6 +640,8 @@ impl DiffState {
             rows: Vec::new(),
             display_rows: Vec::new(),
             display_sides: Vec::new(),
+            source_ranges: Default::default(),
+            lazy_changes: false,
             terminal_rows: [None; 2],
             excerpt_ranges: Vec::new(),
             live_buffer: None,
@@ -1931,8 +1952,15 @@ impl DiffState {
     }
 
     pub(crate) fn source_display_row(&self, focus: Focus, source_row: usize) -> Option<usize> {
-        (0..self.display_rows.len())
+        let ranges = self.source_ranges.get(usize::from(focus == Focus::After))?;
-            .position(|display| self.display_source_row(focus, display) == Some(source_row))
+        if focus == Focus::Files {
+            return None;
+        }
+        let index = ranges.partition_point(|(range, _)| range.end <= source_row);
+        let (range, display) = ranges.get(index)?;
+        range
+            .contains(&source_row)
+            .then(|| display + source_row - range.start)
     }
 
     pub(crate) fn display_focus(&self, display: usize) -> Option<Focus> {
@@ -1979,9 +2007,24 @@ impl DiffState {
         }
     }
 
-    pub(crate) fn source_rows(&self, focus: Focus) -> impl DoubleEndedIterator<Item = usize> + '_ {
+    pub(crate) fn source_rows_in(
-        self.excerpt_ranges.iter().flat_map(move |&(start, end)| {
+        &self,
-            (start..end).filter_map(move |index| self.source_row(focus, index))
+        focus: Focus,
+        start: usize,
+        end: usize,
+    ) -> impl DoubleEndedIterator<Item = usize> + '_ {
+        let ranges = &self.source_ranges[usize::from(focus == Focus::After)];
+        let first = ranges.partition_point(|(range, _)| range.end <= start);
+        let last = ranges
+            .partition_point(|(range, _)| range.start < end)
+            .max(first);
+        ranges[first..last].iter().flat_map(move |(range, _)| {
+            let start = range.start.max(start);
+            start..if focus == Focus::Files {
+                start
+            } else {
+                range.end.min(end)
+            }
         })
     }
 
@@ -2038,36 +2081,53 @@ impl DiffState {
     }
 
     pub(crate) fn corresponding_row(&self, from: Focus, to: Focus, row: usize) -> Option<usize> {
-        let aligned = self.rows.iter().position(|aligned| match from {
+        let aligned = self
-            Focus::Before => aligned.before == Some(row),
+            .source_display_row(from, row)
-            Focus::After => aligned.after == Some(row),
+            .and_then(|display| match self.display_rows[display] {
-            Focus::Files => false,
+                DisplayRow::Line(index) => Some(index),
-        })?;
-        self.display_rows
-            .iter()
-            .enumerate()
-            .filter_map(|(display, row)| match *row {
-                DisplayRow::Line(index) => self
-                    .display_source_row(to, display)
-                    .map(|row| (index.abs_diff(aligned), row)),
                 DisplayRow::Fold { .. } => None,
             })
-            .min_by_key(|&(distance, row)| (distance, row))
+            .or_else(|| {
-            .map(|(_, row)| row)
+                self.rows.iter().position(|aligned| match from {
+                    Focus::Before => aligned.before == Some(row),
+                    Focus::After => aligned.after == Some(row),
+                    Focus::Files => false,
+                })
+            })?;
+        if to == Focus::Files {
+            return None;
+        }
+        let mut nearest = None;
+        for (range, display) in &self.source_ranges[usize::from(to == Focus::After)] {
+            let visible = &self.display_rows[*display..*display + range.len()];
+            let offset = visible
+                .partition_point(|row| matches!(row, DisplayRow::Line(index) if *index < aligned));
+            for offset in [offset.saturating_sub(1), offset.min(visible.len() - 1)] {
+                let DisplayRow::Line(index) = visible[offset] else {
+                    unreachable!()
+                };
+                let candidate = (index.abs_diff(aligned), range.start + offset);
+                if nearest.is_none_or(|previous| candidate < previous) {
+                    nearest = Some(candidate);
+                }
+            }
+        }
+        nearest.map(|(_, row)| row)
     }
 
     pub(crate) fn clamp_source_row(&self, focus: Focus, row: usize) -> Option<usize> {
-        let mut previous = None;
+        let ranges = &self.source_ranges[usize::from(focus == Focus::After)];
-        for visible in self.source_rows(focus) {
+        if focus == Focus::Files {
-            if visible >= row {
+            return None;
-                return Some(match previous {
+        }
-                    Some(previous) if row - previous < visible - row => previous,
+        let index = ranges.partition_point(|(range, _)| range.end <= row);
-                    _ => visible,
+        let next = ranges.get(index).map(|(range, _)| row.max(range.start));
-                });
+        let previous = index.checked_sub(1).map(|index| ranges[index].0.end - 1);
-            }
+        match (previous, next) {
-            previous = Some(visible);
+            (Some(previous), Some(next)) if row - previous < next - row => Some(previous),
+            (_, Some(next)) => Some(next),
+            (previous, None) => previous,
         }
-        previous
     }
 
     fn restore_top(&mut self, anchor: Option<DisplayRow>, side: Option<Focus>) {
@@ -2147,7 +2207,12 @@ impl DiffState {
             old.push("");
         }
         let new = lines.iter().map(String::as_str).collect::<Vec<_>>();
-        let ops = similar::capture_diff_slices(similar::Algorithm::Myers, &old, &new);
+        let ops = similar::capture_diff_slices_deadline(
+            similar::Algorithm::Myers,
+            &old,
+            &new,
+            Some(Instant::now() + Duration::from_millis(100)),
+        );
         let mut mapping = vec![new.len().saturating_sub(1); old.len()];
         for op in &ops {
             let (_, old_range, new_range) = op.as_tag_tuple();
@@ -2257,11 +2322,13 @@ impl DiffState {
         self.terminal_rows = [None; 2];
         self.display_rows.clear();
         self.display_sides.clear();
+        self.source_ranges.iter_mut().for_each(Vec::clear);
         self.excerpt_ranges.clear();
         if reset {
             self.top = 0;
             self.left_before = 0;
             self.left_after = 0;
+            self.lazy_changes = false;
             self.max_before = 0;
             self.max_after = 0;
             self.after_widths.clear();
@@ -2277,6 +2344,7 @@ impl DiffState {
             self.max_after = 0;
             return;
         };
+        self.lazy_changes = file.before.len().saturating_add(file.after.len()) > 20_000;
         if reset {
             self.max_before = file
                 .before
@@ -2387,9 +2455,22 @@ impl DiffState {
         self.excerpt_ranges.truncate(count);
         self.display_rows.clear();
         self.display_sides.clear();
+        self.source_ranges.iter_mut().for_each(Vec::clear);
         if self.excerpt_ranges.is_empty() {
             return;
         }
+        let visible = self
+            .excerpt_ranges
+            .iter()
+            .map(|&(start, end)| end - start)
+            .sum::<usize>();
+        let capacity = visible
+            .saturating_add(self.excerpt_ranges.len())
+            .saturating_add(1);
+        self.display_rows.reserve(capacity);
+        if self.view_mode == ViewMode::Unified {
+            self.display_sides.reserve(capacity);
+        }
         let mut end = 0;
         for &(start, next_end) in &self.excerpt_ranges {
             if end < start {
@@ -2436,6 +2517,27 @@ impl DiffState {
                 self.display_sides.push(None);
             }
         }
+        self.rebuild_source_ranges();
+    }
+
+    fn rebuild_source_ranges(&mut self) {
+        self.source_ranges.iter_mut().for_each(Vec::clear);
+        for display in 0..self.display_rows.len() {
+            for (side, focus) in [Focus::Before, Focus::After].into_iter().enumerate() {
+                let Some(source) = self.display_source_row(focus, display) else {
+                    continue;
+                };
+                let ranges = &mut self.source_ranges[side];
+                if let Some((range, first)) = ranges.last_mut()
+                    && range.end == source
+                    && *first + range.len() == display
+                {
+                    range.end += 1;
+                } else {
+                    ranges.push((source..source + 1, display));
+                }
+            }
+        }
     }
 
     pub(crate) fn expand_source_excerpt(&mut self, focus: Focus, below: bool, source_row: usize) {
@@ -2526,27 +2628,55 @@ fn refresh_character_changes(
     file: &ChangedFile,
     after_rows: std::ops::Range<usize>,
 ) {
+    let deadline = Instant::now() + Duration::from_millis(20);
     for row in rows {
         if row.kind == RowKind::Changed
             && let Some(after) = row.after.filter(|index| after_rows.contains(index))
         {
-            row.changes = character_changes(
+            if file.before.len().saturating_add(file.after.len()) <= 20_000 {
-                &file.before[row.before.expect("paired changed row")],
+                row.changes = character_changes_deadline(
-                &file.after[after],
+                    &file.before[row.before.expect("paired changed row")],
-            );
+                    &file.after[after],
+                    deadline,
+                );
+            } else {
+                row.changes.clear();
+            }
         }
     }
 }
 
 fn character_changes(before: &str, after: &str) -> Vec<[std::ops::Range<usize>; 2]> {
+    character_changes_deadline(before, after, Instant::now() + Duration::from_millis(2))
+}
+
+fn character_changes_deadline(
+    before: &str,
+    after: &str,
+    deadline: Instant,
+) -> Vec<[std::ops::Range<usize>; 2]> {
     if before == after {
         return Vec::new();
     }
+    // Large/minified lines keep their complete text and line-level change shading,
+    // without allocating a grapheme table proportional to their size.
+    if before.len().saturating_add(after.len()) > 64 * 1024 || Instant::now() >= deadline {
+        return vec![[0..usize::MAX, 0..usize::MAX]];
+    }
     let before: Vec<_> = before.graphemes(true).collect();
     let after: Vec<_> = after.graphemes(true).collect();
     let mut changes: Vec<[std::ops::Range<usize>; 2]> = Vec::new();
     let (mut before_offset, mut after_offset) = (0, 0);
-    for op in similar::capture_diff_slices(similar::Algorithm::Myers, &before, &after) {
+    for op in similar::capture_diff_slices_deadline(
+        similar::Algorithm::Myers,
+        &before,
+        &after,
+        if before.len().saturating_add(after.len()) <= 2048 {
+            None
+        } else {
+            Some(deadline.min(Instant::now() + Duration::from_millis(2)))
+        },
+    ) {
         let (tag, before, after) = op.as_tag_tuple();
         // similar's optimized empty-range anchors can be stale; consume lengths in order.
         let before = before_offset..before_offset + before.len();
@@ -2626,6 +2756,19 @@ fn short_match_between_large_changes(
 fn line_width(line: &str) -> usize {
     #[cfg(test)]
     tests::WIDTH_SCANS.with(|scans| scans.set(scans.get() + 1));
+    if line.is_ascii() {
+        if !line.bytes().any(|byte| byte < 32 || byte == 127) {
+            return line.len();
+        }
+        return line.bytes().fold(0, |column, byte| {
+            column
+                + match byte {
+                    b'\t' => 4 - column % 4,
+                    0..=31 | 127 => 0,
+                    _ => 1,
+                }
+        });
+    }
     line.graphemes(true).fold(0, |column, grapheme| {
         column
             + if grapheme == "\t" {
@@ -2720,27 +2863,82 @@ pub(crate) fn line_hunks(
     after: &[String],
     after_eof: bool,
 ) -> Vec<Hunk> {
-    fn tokens(lines: &[String], eof: bool) -> Vec<(&str, bool)> {
+    // Empty sides need neither a token table nor a Myers frontier.
-        lines
+    if before.is_empty() || after.is_empty() {
-            .iter()
+        return if before.is_empty() && after.is_empty() {
-            .enumerate()
+            Vec::new()
-            .map(|(index, line)| (line.as_str(), index + 1 < lines.len() || eof))
+        } else {
+            vec![Hunk {
+                before_start: 0,
+                before_len: before.len(),
+                after_start: 0,
+                after_len: after.len(),
+            }]
+        };
+    }
+    let same = |old: usize, new: usize| {
+        before[old] == after[new]
+            && (old + 1 < before.len() || before_eof) == (new + 1 < after.len() || after_eof)
+    };
+    let large = before.len().saturating_add(after.len()) > 20_000;
+    // Keep ordinary Myers/Compact anchoring unchanged; large comparisons avoid
+    // tokenizing their already-equal prefix and suffix.
+    let (start, suffix) = if large {
+        let start = (0..before.len().min(after.len()))
+            .take_while(|&index| same(index, index))
+            .count();
+        let suffix = (0..(before.len() - start).min(after.len() - start))
+            .take_while(|&offset| same(before.len() - 1 - offset, after.len() - 1 - offset))
+            .count();
+        (start, suffix)
+    } else {
+        (0, 0)
+    };
+    let before_end = before.len() - suffix;
+    let after_end = after.len() - suffix;
+    if start == before_end || start == after_end {
+        return if start == before_end && start == after_end {
+            Vec::new()
+        } else {
+            vec![Hunk {
+                before_start: start,
+                before_len: before_end - start,
+                after_start: start,
+                after_len: after_end - start,
+            }]
+        };
+    }
+    fn tokens(lines: &[String], eof: bool, range: std::ops::Range<usize>) -> Vec<(&str, bool)> {
+        range
+            .map(|index| (lines[index].as_str(), index + 1 < lines.len() || eof))
             .collect()
     }
-    let before = tokens(before, before_eof);
+    let old = tokens(before, before_eof, start..before_end);
-    let after = tokens(after, after_eof);
+    let new = tokens(after, after_eof, start..after_end);
-    similar::capture_diff_slices(similar::Algorithm::Myers, &before, &after)
+    // ponytail: after 100ms Myers may conservatively replace unresolved lines;
-        .into_iter()
+    // every source line remains present, never mislabeled as equal.
-        .filter_map(|op| {
+    let mut old_offset = start;
-            let (tag, before, after) = op.as_tag_tuple();
+    let mut new_offset = start;
-            (tag != similar::DiffTag::Equal).then_some(Hunk {
+    similar::capture_diff_slices_deadline(
-                before_start: before.start,
+        similar::Algorithm::Myers,
-                before_len: before.len(),
+        &old,
-                after_start: after.start,
+        &new,
-                after_len: after.len(),
+        Some(Instant::now() + Duration::from_millis(100)),
-            })
+    )
-        })
+    .into_iter()
-        .collect()
+    .filter_map(|op| {
+        let (tag, old, new) = op.as_tag_tuple();
+        let hunk = Hunk {
+            before_start: if large { old_offset } else { old.start },
+            before_len: old.len(),
+            after_start: if large { new_offset } else { new.start },
+            after_len: new.len(),
+        };
+        old_offset += old.len();
+        new_offset += new.len();
+        (tag != similar::DiffTag::Equal).then_some(hunk)
+    })
+    .collect()
 }
 
 fn file_is_changed(file: &ChangedFile) -> bool {

src/editor.rs

@@ -11797,8 +11797,7 @@ impl Editor {
             .and_then(|focus| self.diff().map(|diff| (diff, focus)));
         (start..if diff.is_none() { end } else { start }).chain(
             diff.into_iter()
-                .flat_map(|(diff, focus)| diff.source_rows(focus))
+                .flat_map(move |(diff, focus)| diff.source_rows_in(focus, start, end)),
-                .filter(move |row| start <= *row && *row < end),
         )
     }
 

src/repository.rs

@@ -21,6 +21,10 @@ use std::{
 
 const OUTPUT_LIMIT: usize = 64 * 1024 * 1024;
 const FILE_LIMIT: usize = 16 * 1024 * 1024;
+// Comparison payloads are independent of picker/gutter annotation budgets.
+const COMPARISON_FILE_LIMIT: usize = 64 * 1024 * 1024;
+const COMPARISON_LINE_LIMIT: usize = 4_000_000;
+const WORKING_SNAPSHOT_LIMIT: usize = 512 * 1024 * 1024;
 const TIMEOUT: Duration = Duration::from_secs(60);
 static NEXT_TEMP: AtomicU64 = AtomicU64::new(0);
 
@@ -1948,94 +1952,54 @@ impl Repository {
     }
 
     fn materialize(&self, snapshot: &mut Snapshot) -> Result<(), String> {
-        let stores: BTreeSet<_> = snapshot
+        for (path, file) in snapshot.iter_mut() {
-            .values()
+            self.check_cancelled()?;
-            .filter(|file| file.projection.is_none())
+            if let Some((root, id)) = &file.projection {
-            .filter_map(|file| file.object.as_ref().map(|(store, _)| store.clone()))
+                let revset = format!("commit_id({})", serde_json::to_string(id).unwrap());
-            .collect();
+                let pattern =
-        for store in stores {
+                    jj_file_pattern(path.to_str().ok_or("Non-UTF-8 Jujutsu conflict path")?);
-            let mut request = Vec::new();
+                file.bytes = self
-            for file in snapshot.values() {
+                    .jj(
-                if let Some((source, id)) = &file.object
+                        root,
-                    && source == &store
+                        &["file", "show", "-r", &revset, "-T", "", "--", &pattern],
-                    && file.projection.is_none()
+                    )
-                {
+                    .map_err(|error| format!("{}: {error}", path.display()))?;
-                    request.extend_from_slice(id.as_bytes());
+            } else if let Some((store, id)) = &file.object {
-                    request.push(b'\n');
+                let size = text(self.git(store, &["cat-file", "-s", id])?)?
-                }
+                    .trim()
-            }
+                    .parse::<usize>()
-            let mut cmd = git_command(&store);
-            cmd.args(["cat-file", "--batch"]);
-            let data = checked(run(cmd, request, &self.cancelled)?)?;
-            let mut offset = 0;
-            for (path, file) in snapshot.iter_mut() {
-                if file.projection.is_some()
-                    || file
-                        .object
-                        .as_ref()
-                        .is_none_or(|(source, _)| source != &store)
-                {
-                    continue;
-                }
-                self.check_cancelled()?;
-                let end = offset
-                    + data[offset..]
-                        .iter()
-                        .position(|b| *b == b'\n')
-                        .ok_or("Truncated Git object header")?;
-                let header = std::str::from_utf8(&data[offset..end])
-                    .map_err(|_| "Invalid Git object header")?;
-                let size: usize = header
-                    .rsplit(' ')
-                    .next()
-                    .ok_or("Missing Git object size")?
-                    .parse()
                     .map_err(|_| "Invalid Git object size")?;
-                if size > FILE_LIMIT {
+                if size > COMPARISON_FILE_LIMIT {
                     return Err(format!(
-                        "{} exceeds the 16 MiB comparison file limit",
+                        "{} exceeds the 64 MiB per-side comparison file limit",
                         path.display()
                     ));
                 }
-                offset = end + 1;
+                // One blob per subprocess output: never concatenate both sides or
-                if size >= data.len().saturating_sub(offset) {
+                // copy a cat-file batch into a second allocation.
-                    return Err("Truncated Git object content".into());
+                file.bytes = self.git(store, &["cat-file", "blob", id])?;
-                }
+            } else {
-                file.bytes = data[offset..offset + size].to_vec();
-                file.object = None;
-                offset += size + 1;
-            }
-        }
-        let mut total = 0;
-        for (path, file) in snapshot.iter_mut() {
-            let Some((root, id)) = &file.projection else {
                 continue;
-            };
+            }
-            let revset = format!("commit_id({})", serde_json::to_string(id).unwrap());
+            if file.bytes.len() > COMPARISON_FILE_LIMIT {
-            let pattern = jj_file_pattern(path.to_str().ok_or("Non-UTF-8 Jujutsu conflict path")?);
-            let bytes = self.jj(
-                root,
-                &["file", "show", "-r", &revset, "-T", "", "--", &pattern],
-            )?;
-            if bytes.len() > FILE_LIMIT {
                 return Err(format!(
-                    "{} exceeds the 16 MiB comparison file limit",
+                    "{} exceeds the 64 MiB per-side comparison file limit",
                     path.display()
                 ));
             }
-            total += bytes.len();
-            if total > OUTPUT_LIMIT {
-                return Err(
-                    "Jujutsu conflict projection exceeds the 64 MiB comparison limit".into(),
-                );
-            }
-            file.bytes = bytes;
             file.object = None;
             file.projection = None;
         }
         Ok(())
     }
-    fn working_snapshot(&self, buffers: &[(PathBuf, String)]) -> Result<Snapshot, String> {
+    fn working_snapshot(&mut self, buffers: &[(PathBuf, String)]) -> Result<Snapshot, String> {
+        self.ensure_private()?;
+        let store = self.private.as_ref().unwrap().1.clone();
+        // Freeze each file once, then let Git ingest them in one batch. Keeping
+        // object identities makes catalog comparison lazy even against Working.
+        let frozen = Temporary::new()?;
+        let mut frozen_paths = Vec::new();
+        let mut hash_input = Vec::new();
         let mut tracked = BTreeSet::new();
         let mut submodules = BTreeSet::new();
         if self.kind == RepoKind::Git {
@@ -2220,39 +2184,46 @@ impl Repository {
                 let bytes = target.to_string_lossy().as_bytes().to_vec();
                 ("120000".into(), bytes, None)
             } else if metadata.is_file() {
-                if metadata.len() > FILE_LIMIT as u64 {
+                if metadata.len() > COMPARISON_FILE_LIMIT as u64 {
                     return Err(format!(
-                        "{} exceeds the 16 MiB comparison file limit",
+                        "{} exceeds the 64 MiB per-side comparison file limit",
                         path.display()
                     ));
                 }
-                let bytes = fs::File::open(&full)
+                let mut source = fs::File::open(&full)
-                    .and_then(|f| {
+                    .map_err(|e| format!("{}: {e}", path.display()))?
-                        let mut bytes = Vec::new();
+                    .take(COMPARISON_FILE_LIMIT as u64 + 1);
-                        f.take(FILE_LIMIT as u64 + 1).read_to_end(&mut bytes)?;
+                let mut bytes = Vec::with_capacity(metadata.len() as usize);
-                        Ok(bytes)
+                let mut chunk = [0; 64 * 1024];
-                    })
+                loop {
-                    .map_err(|e| format!("{}: {e}", path.display()))?;
+                    self.check_cancelled()?;
-                if bytes.len() > FILE_LIMIT {
+                    let count = source
-                    return Err(format!(
+                        .read(&mut chunk)
-                        "{} grew beyond the 16 MiB comparison file limit",
+                        .map_err(|e| format!("{}: {e}", path.display()))?;
-                        path.display()
+                    if count == 0 {
-                    ));
+                        break;
-                }
+                    }
-                let bytes = if !bytes.contains(&0)
+                    if bytes.len() + count > COMPARISON_FILE_LIMIT {
-                    && std::str::from_utf8(&bytes).is_ok()
-                    && let Some(text) = overlays.get(&path)
-                {
-                    if text.len() > FILE_LIMIT {
                         return Err(format!(
-                            "{} exceeds the 16 MiB comparison file limit",
+                            "{} grew beyond the 64 MiB per-side comparison file limit",
                             path.display()
                         ));
                     }
-                    text.as_bytes().to_vec()
+                    bytes.extend_from_slice(&chunk[..count]);
-                } else {
+                }
-                    bytes
+                if !bytes.contains(&0)
-                };
+                    && std::str::from_utf8(&bytes).is_ok()
+                    && let Some(text) = overlays.get(&path)
+                {
+                    if text.len() > COMPARISON_FILE_LIMIT {
+                        return Err(format!(
+                            "{} exceeds the 64 MiB per-side comparison file limit",
+                            path.display()
+                        ));
+                    }
+                    bytes.clear();
+                    bytes.extend_from_slice(text.as_bytes());
+                }
                 #[cfg(unix)]
                 let executable = {
                     use std::os::unix::fs::PermissionsExt;
@@ -2273,20 +2244,45 @@ impl Repository {
                 )
             };
             total += bytes.len();
-            if total > OUTPUT_LIMIT {
+            if total > WORKING_SNAPSHOT_LIMIT {
-                return Err("Working snapshot exceeds the 64 MiB comparison limit".into());
+                return Err("Working snapshot exceeds the 512 MiB comparison limit".into());
             }
+            let object = if notice.is_none() {
+                let name = frozen_paths.len().to_string();
+                fs::write(frozen.0.join(&name), &bytes)
+                    .map_err(|e| format!("{}: {e}", path.display()))?;
+                writeln!(&mut hash_input, "{name}").unwrap();
+                frozen_paths.push(path.clone());
+                Some((store.clone(), String::new()))
+            } else {
+                None
+            };
             snapshot.insert(
                 path,
                 FileData {
                     mode,
-                    bytes,
+                    bytes: vec![],
-                    object: None,
+                    object,
                     projection: None,
                     notice,
                 },
             );
         }
+        if !frozen_paths.is_empty() {
+            let mut hash = git_command(&store);
+            hash.args(["hash-object", "--no-filters", "-w", "--stdin-paths"])
+                .current_dir(&frozen.0);
+            let hashes = text(checked(run(hash, hash_input, &self.cancelled)?)?)?;
+            let mut hashes = hashes.lines();
+            for path in frozen_paths {
+                self.check_cancelled()?;
+                let id = hashes.next().ok_or("Truncated Working snapshot hashes")?;
+                snapshot.get_mut(&path).unwrap().object = Some((store.clone(), id.into()));
+            }
+            if hashes.next().is_some() {
+                return Err("Unexpected Working snapshot hashes".into());
+            }
+        }
         Ok(snapshot)
     }
     fn snapshot_revision(
@@ -2299,13 +2295,16 @@ impl Repository {
         }
         self.ensure_private()?;
         let store = &self.private.as_ref().unwrap().1;
-        let mut input = b"vamp-working-snapshot-v1\0".to_vec();
+        let mut input = b"vamp-working-snapshot-v2\0".to_vec();
         for (path, file) in snapshot {
             self.check_cancelled()?;
             for field in [
                 path.as_os_str().as_encoded_bytes(),
                 file.mode.as_bytes(),
-                file.bytes.as_slice(),
+                file.object
+                    .as_ref()
+                    .map_or("", |(_, id)| id.as_str())
+                    .as_bytes(),
                 file.notice.as_deref().unwrap_or("").as_bytes(),
             ] {
                 input.extend_from_slice(&(field.len() as u64).to_be_bytes());
@@ -2760,15 +2759,6 @@ impl Repository {
     ) -> Result<Comparison, String> {
         let left = &mut prepared.left;
         let right = &mut prepared.right;
-        // Frozen checkout bytes have no object identity. Compare them in a single
-        // batch; immutable blobs and deferred conflict projections retain IDs.
-        let frozen = |file: &FileData| {
-            file.object.is_none() && matches!(file.mode.as_str(), "100644" | "100755" | "120000")
-        };
-        if !same && (left.values().any(frozen) || right.values().any(frozen)) {
-            self.materialize(left)?;
-            self.materialize(right)?;
-        }
         let right_view = if same { &*left } else { &*right };
         let mut files = vec![];
         let paths: BTreeSet<_> = left.keys().chain(right_view.keys()).collect();
@@ -2950,6 +2940,19 @@ impl Repository {
             notices.push("Binary or non-UTF-8 content; text diff is unavailable".into());
             (vec![], vec![], vec![])
         } else {
+            // Bound line-header allocations as well as payload bytes. The real
+            // generated Honey parser has more than 1.5 million lines.
+            for bytes in [a, b] {
+                let count = bytes.iter().filter(|&&byte| byte == b'\n').count()
+                    + usize::from(!bytes.is_empty() && !bytes.ends_with(b"\n"));
+                if count > COMPARISON_LINE_LIMIT {
+                    return Err(format!(
+                        "{} exceeds the 4000000-line per-side comparison limit",
+                        path.display()
+                    ));
+                }
+                self.check_cancelled()?;
+            }
             if (!a.is_empty() && !a.ends_with(b"\n")) || (!b.is_empty() && !b.ends_with(b"\n")) {
                 notices.push(format!(
                     "EOF newline: Before {}, After {}",

src/tui.rs

@@ -3976,13 +3976,21 @@ impl WheelPanels {
         } else {
             let content = diff_pane_panel(frame, diff, selected, panes, true, resolver);
             frame.render_widget(
-                Paragraph::new(if diff.loading {
+                Paragraph::new(display_text(if diff.loading {
                     "Loading file…"
+                } else if let Some(error) = diff.error.as_deref() {
+                    error
                 } else if selected.is_some() {
                     "File details unavailable · select the file to retry"
                 } else {
                     "No changed files"
-                }),
+                }))
+                .wrap(Wrap { trim: false })
+                .style(Style::default().fg(if diff.error.is_some() {
+                    Color::Red
+                } else {
+                    Color::Reset
+                })),
                 content,
             );
         }

tests/repository.rs

@@ -1728,18 +1728,30 @@ fn conflicted_catalog_preserves_symlinks_and_defers_unselected_contents() {
     let fixture = Fixture::new();
     let root = fixture.dir("lazy-conflict");
     init_jj(&root, false);
-    fs::write(root.join("conflict"), "base\n").unwrap();
+    jj(
+        &root,
+        &[
+            "config",
+            "set",
+            "--repo",
+            "snapshot.max-new-file-size",
+            "100MiB",
+        ],
+    );
+    let tail_line = "x".repeat(1023);
+    let tail = format!("{tail_line}\n").repeat(17 * 1024 + 1);
+    fs::write(root.join("conflict"), format!("base\n{tail}")).unwrap();
     fs::write(root.join("unselected"), "old\n").unwrap();
     symlink("unselected", root.join("unchanged-link")).unwrap();
     jj(&root, &["describe", "-m", "base"]);
     let base = jj(&root, &["log", "--no-graph", "-r", "@", "-T", "commit_id"]);
     jj(&root, &["new", &base]);
-    fs::write(root.join("conflict"), "left\n").unwrap();
+    fs::write(root.join("conflict"), format!("left\n{tail}")).unwrap();
     fs::write(root.join("unselected"), vec![b'x'; 17 * 1024 * 1024]).unwrap();
     jj(&root, &["describe", "-m", "left"]);
     let left = jj(&root, &["log", "--no-graph", "-r", "@", "-T", "commit_id"]);
     jj(&root, &["new", &base]);
-    fs::write(root.join("conflict"), "right\n").unwrap();
+    fs::write(root.join("conflict"), format!("right\n{tail}")).unwrap();
     jj(&root, &["describe", "-m", "right"]);
     let right = jj(&root, &["log", "--no-graph", "-r", "@", "-T", "commit_id"]);
     jj(&root, &["new", &left, &right, "-m", "conflicted merge"]);
@@ -1764,7 +1776,9 @@ fn conflicted_catalog_preserves_symlinks_and_defers_unselected_contents() {
         [Path::new("conflict"), Path::new("unselected")]
     );
     let conflict = repository.load_file(1, 0).unwrap();
-    assert_eq!(conflict.before, ["base"]);
+    assert_eq!(conflict.before[0], "base");
+    assert_eq!(conflict.before.len(), 17 * 1024 + 2);
+    assert!(conflict.before[1..].iter().all(|line| line == &tail_line));
     let projection = jj(
         &root,
         &[
@@ -1779,9 +1793,13 @@ fn conflicted_catalog_preserves_symlinks_and_defers_unselected_contents() {
             "conflict",
         ],
     );
+    assert!(projection.len() > 16 * 1024 * 1024);
     assert_eq!(conflict.after, projection.lines().collect::<Vec<_>>());
     assert!(!conflict.hunks.is_empty());
-    assert!(repository.load_file(1, 1).is_err());
+    let large = repository.load_file(1, 1).unwrap();
+    assert_eq!(large.after.len(), 1);
+    assert_eq!(large.after[0].len(), 17 * 1024 * 1024);
+    assert!(large.after[0].bytes().all(|byte| byte == b'x'));
     assert_eq!(disk_state(&root), state);
 
     jj(&root, &["new", &merged]);
@@ -1945,6 +1963,220 @@ fn immutable_catalog_skips_unchanged_blobs_and_prepares_only_requested_details()
     assert!(repository.load_file(100, index).is_err());
 }
 
+#[test]
+fn comparison_size_and_line_ceilings_are_lazy_and_explicit() {
+    let fixture = Fixture::new();
+    let root = fixture.dir("ceilings");
+    init_git(&root);
+    git(&root, &["commit", "--allow-empty", "-m", "before"]);
+    let before = revision(&git(&root, &["rev-parse", "HEAD"]), RepoKind::Git);
+    fs::File::create(root.join("too-many-bytes"))
+        .unwrap()
+        .set_len(64 * 1024 * 1024 + 1)
+        .unwrap();
+    fs::write(root.join("too-many-lines"), vec![b'\n'; 4_000_001]).unwrap();
+    fs::write(root.join("selected"), "still available\n").unwrap();
+    git(&root, &["add", "."]);
+    git(&root, &["commit", "-m", "after"]);
+    let after = revision(&git(&root, &["rev-parse", "HEAD"]), RepoKind::Git);
+    let mut repository = open(&root);
+    let catalog = repository
+        .compare(1, &before, &after, ComparisonMode::Snapshots, &[])
+        .unwrap();
+    assert_eq!(catalog.files.len(), 3);
+    assert!(catalog.files.iter().all(|file| !file.loaded));
+    for (index, entry) in catalog.files.iter().enumerate() {
+        match entry.path.to_str().unwrap() {
+            "selected" => assert_eq!(
+                repository.load_file(1, index).unwrap().after,
+                ["still available"]
+            ),
+            "too-many-bytes" => assert!(
+                repository
+                    .load_file(1, index)
+                    .unwrap_err()
+                    .contains("64 MiB per-side comparison file limit")
+            ),
+            "too-many-lines" => assert!(
+                repository
+                    .load_file(1, index)
+                    .unwrap_err()
+                    .contains("4000000-line per-side comparison limit")
+            ),
+            path => panic!("Unexpected file {path}"),
+        }
+    }
+}
+
+#[test]
+fn large_git_and_jj_comparisons_load_exact_snapshots_and_frozen_working_overlays() {
+    let fixture = Fixture::new();
+    // Both modified sides exceed 52.3 MB; the Working pair exceeds the old
+    // aggregate 64 MiB budget as well as its old 16 MiB per-file limit.
+    let body_line = "x".repeat(1023);
+    let mut contents = format!(
+        "before\n{}",
+        format!("{body_line}\n").repeat(52_300_000 / 1024 + 1)
+    );
+    let line_count = contents.lines().count();
+    for kind in [RepoKind::Git, RepoKind::Jujutsu] {
+        let root = fixture
+            .dir(if kind == RepoKind::Git { "git" } else { "jj" })
+            .canonicalize()
+            .unwrap();
+        match kind {
+            RepoKind::Git => init_git(&root),
+            RepoKind::Jujutsu => {
+                init_jj(&root, false);
+                jj(
+                    &root,
+                    &[
+                        "config",
+                        "set",
+                        "--repo",
+                        "snapshot.max-new-file-size",
+                        "100MiB",
+                    ],
+                );
+            }
+        }
+        let set_header = |contents: &mut String, header: &str| {
+            let end = contents.find('\n').unwrap();
+            contents.replace_range(..end, header);
+        };
+        set_header(&mut contents, "before");
+        fs::write(root.join("modified"), &contents).unwrap();
+        set_header(&mut contents, "removed");
+        fs::write(root.join("removed"), &contents).unwrap();
+        let before = match kind {
+            RepoKind::Git => {
+                git(&root, &["add", "."]);
+                git(&root, &["commit", "-m", "large before"]);
+                git(&root, &["rev-parse", "HEAD"])
+            }
+            RepoKind::Jujutsu => {
+                jj(&root, &["describe", "-m", "large before"]);
+                let id = jj(&root, &["log", "--no-graph", "-r", "@", "-T", "commit_id"]);
+                jj(&root, &["new", &id]);
+                id
+            }
+        };
+        set_header(&mut contents, "after");
+        fs::write(root.join("modified"), &contents).unwrap();
+        fs::remove_file(root.join("removed")).unwrap();
+        set_header(&mut contents, "added");
+        fs::write(root.join("added"), &contents).unwrap();
+        let after = match kind {
+            RepoKind::Git => {
+                git(&root, &["add", "."]);
+                git(&root, &["commit", "-m", "large after"]);
+                git(&root, &["rev-parse", "HEAD"])
+            }
+            RepoKind::Jujutsu => {
+                jj(&root, &["describe", "-m", "large after"]);
+                jj(&root, &["log", "--no-graph", "-r", "@", "-T", "commit_id"])
+            }
+        };
+        let assert_contents = |lines: &[String], header: &str| {
+            assert_eq!(lines.len(), line_count);
+            assert_eq!(lines[0], header);
+            assert!(lines[1..].iter().all(|line| line == &body_line));
+        };
+        let cancelled = Arc::new(AtomicBool::new(false));
+        let mut repository = Repository::open(&root, cancelled.clone()).unwrap();
+        let catalog = repository
+            .compare(
+                1,
+                &revision(&before, kind),
+                &revision(&after, kind),
+                ComparisonMode::Snapshots,
+                &[],
+            )
+            .unwrap();
+        assert_eq!(catalog.files.len(), 3);
+        assert!(catalog.files.iter().all(|file| !file.loaded
+            && file.before.is_empty()
+            && file.after.is_empty()
+            && file.hunks.is_empty()));
+        for (index, entry) in catalog.files.iter().enumerate() {
+            let file = repository.load_file(1, index).unwrap();
+            assert!(file.loaded);
+            assert_eq!(file.hunks.len(), 1);
+            let hunk = &file.hunks[0];
+            let expected = match entry.path.to_str().unwrap() {
+                "added" => {
+                    assert_eq!(file.status, FileStatus::Added);
+                    assert!(file.before.is_empty());
+                    assert_contents(&file.after, "added");
+                    (0, 0, 0, line_count)
+                }
+                "removed" => {
+                    assert_eq!(file.status, FileStatus::Removed);
+                    assert_contents(&file.before, "removed");
+                    assert!(file.after.is_empty());
+                    (0, line_count, 0, 0)
+                }
+                "modified" => {
+                    assert_eq!(file.status, FileStatus::Modified);
+                    assert_contents(&file.before, "before");
+                    assert_contents(&file.after, "after");
+                    (0, 1, 0, 1)
+                }
+                path => panic!("Unexpected file {path}"),
+            };
+            assert_eq!(
+                (
+                    hunk.before_start,
+                    hunk.before_len,
+                    hunk.after_start,
+                    hunk.after_len
+                ),
+                expected
+            );
+            assert!(file.before.is_empty() || file.before_eof);
+            assert!(file.after.is_empty() || file.after_eof);
+        }
+        set_header(&mut contents, "overlay");
+        let buffers = [(root.join("modified"), contents)];
+        let working = repository
+            .compare(
+                2,
+                &revision(&after, kind),
+                &working(kind),
+                ComparisonMode::Snapshots,
+                &buffers,
+            )
+            .unwrap();
+        assert_eq!(working.files.len(), 1);
+        assert!(!working.files[0].loaded);
+        assert!(working.editable_paths.contains(Path::new("modified")));
+        fs::write(root.join("modified"), "later disk contents\n").unwrap();
+        cancelled.store(true, Ordering::Release);
+        assert!(
+            repository
+                .load_file(2, 0)
+                .unwrap_err()
+                .contains("cancelled")
+        );
+        cancelled.store(false, Ordering::Release);
+        let file = repository.load_file(2, 0).unwrap();
+        assert_contents(&file.before, "after");
+        assert_contents(&file.after, "overlay");
+        assert_eq!(file.hunks.len(), 1);
+        let hunk = &file.hunks[0];
+        assert_eq!(
+            (
+                hunk.before_start,
+                hunk.before_len,
+                hunk.after_start,
+                hunk.after_len
+            ),
+            (0, 1, 0, 1)
+        );
+        contents = buffers.into_iter().next().unwrap().1;
+    }
+}
+
 #[test]
 fn deferred_file_details_keep_frozen_checkout_and_unsaved_after() {
     let fixture = Fixture::new();
@@ -2112,8 +2344,16 @@ fn working_after_refresh_preserves_frozen_working_before_and_sources_on_failure(
     assert!(repository.refresh_checkout(10, 11, true, &[]).is_err());
     cancelled.store(false, Ordering::Release);
     assert_eq!(repository.load_file(10, 0).unwrap().after, ["frozen after"]);
-    fs::write(root.join("oversized"), vec![b'x'; 17 * 1024 * 1024]).unwrap();
+    fs::File::create(root.join("oversized"))
-    assert!(repository.refresh_checkout(10, 11, true, &[]).is_err());
+        .unwrap()
+        .set_len(64 * 1024 * 1024 + 1)
+        .unwrap();
+    assert!(
+        repository
+            .refresh_checkout(10, 11, true, &[])
+            .unwrap_err()
+            .contains("64 MiB per-side comparison file limit")
+    );
     let original = repository.load_file(10, 0).unwrap();
     assert_eq!(original.before, ["frozen before"]);
     assert_eq!(original.after, ["frozen after"]);
@@ -2194,7 +2434,7 @@ fn working_after_refresh_preserves_frozen_working_before_and_sources_on_failure(
 }
 
 #[test]
-fn refresh_materialization_failure_restores_prepared_file_sources() {
+fn refresh_keeps_historical_blob_sources_lazy_even_with_working_files() {
     let fixture = Fixture::new();
     let root = fixture.dir("failed-materialization");
     init_git(&root);
@@ -2206,7 +2446,7 @@ fn refresh_materialization_failure_restores_prepared_file_sources() {
     fs::remove_file(root.join("a-file")).unwrap();
     fs::remove_file(root.join("z-large")).unwrap();
     let mut repository = open(&root);
-    let original = repository
+    repository
         .compare(
             1,
             &base,
@@ -2215,20 +2455,29 @@ fn refresh_materialization_failure_restores_prepared_file_sources() {
             &[],
         )
         .unwrap();
-    let index = original
+    assert_eq!(
+        repository.load_file(1, 0).unwrap().before,
+        ["frozen source"]
+    );
+    fs::write(root.join("a-file"), "new checkout\n").unwrap();
+    git(&root, &["add", "."]);
+    git(&root, &["commit", "-m", "descendant"]);
+    let refreshed = repository
+        .refresh_checkout(1, 2, true, &[])
+        .unwrap()
+        .map(checkout_comparison)
+        .unwrap();
+    assert!(refreshed.files.iter().all(|file| !file.loaded));
+    let index = refreshed
         .files
         .iter()
         .position(|file| file.path == Path::new("a-file"))
         .unwrap();
-    fs::write(root.join("a-file"), "new checkout\n").unwrap();
+    let file = repository.load_file(2, index).unwrap();
-    git(&root, &["add", "."]);
-    git(&root, &["commit", "-m", "descendant"]);
-    assert!(repository.refresh_checkout(1, 2, true, &[]).is_err());
-    let file = repository.load_file(1, index).unwrap();
     assert_eq!(file.before, ["frozen source"]);
-    assert!(file.after.is_empty());
+    assert_eq!(file.after, ["new checkout"]);
-    assert_eq!(file.status, FileStatus::Removed);
+    assert_eq!(file.status, FileStatus::Modified);
-    assert!(repository.load_file(2, index).is_err());
+    assert!(repository.load_file(1, index).is_err());
 }
 
 #[test]

tests/unit/diff.rs

@@ -2240,7 +2240,7 @@ fn unified_groups_changed_runs_and_roundtrips_side_specific_source_rows() {
     );
     assert!(state.after_editable());
     for focus in [Focus::Before, Focus::After] {
-        for source in state.source_rows(focus) {
+        for source in state.source_rows_in(focus, 0, usize::MAX) {
             let display = state.source_display_row(focus, source).unwrap();
             assert_eq!(state.display_source_row(focus, display), Some(source));
         }
@@ -2271,10 +2271,15 @@ fn unified_live_reflow_preserves_removed_top_and_expanded_source_excerpts() {
     assert_eq!(state.display_source_row(Focus::Before, state.top), Some(20));
     assert!(state.source_display_row(Focus::After, 17).is_some());
     state.reveal_source_row(Focus::After, 35);
-    let visible: Vec<_> = state.source_rows(Focus::After).collect();
+    let visible: Vec<_> = state.source_rows_in(Focus::After, 0, usize::MAX).collect();
     state.set_view_mode(ViewMode::Split);
     state.set_view_mode(ViewMode::Unified);
-    assert_eq!(state.source_rows(Focus::After).collect::<Vec<_>>(), visible);
+    assert_eq!(
+        state
+            .source_rows_in(Focus::After, 0, usize::MAX)
+            .collect::<Vec<_>>(),
+        visible
+    );
     assert_eq!(state.display_source_row(Focus::After, state.top), Some(35));
 }
 
@@ -2750,3 +2755,251 @@ fn review_anchors_match_frozen_context_and_exact_live_eof_without_allocating() {
     state.comparison.as_mut().unwrap().revisions[0] = "other".into();
     assert!(!state.review_anchor_current(&before));
 }
+
+#[test]
+fn million_row_additions_and_removals_keep_complete_sources_and_compact_navigation() {
+    let lines = vec!["static const unsigned int parser_state = 1234567890;".to_owned(); 1_000_000];
+    let mut state = live_state(Vec::new(), lines, true);
+    state.comparison.as_mut().unwrap().files[0].status = FileStatus::Added;
+    state.align_file();
+    assert_eq!(state.rows.len(), 1_000_000);
+    assert!(state.rows.iter().all(|row| row.changes.is_empty()));
+    assert_eq!(state.source_ranges[1], [(0..1_000_000, 0)]);
+    assert_eq!(state.after_display_row(999_999), Some(999_999));
+    assert_eq!(
+        state.clamp_source_row(Focus::After, usize::MAX),
+        Some(999_999)
+    );
+    let mut visible = state.source_rows_in(Focus::After, 999_990, 999_999);
+    assert_eq!(visible.next(), Some(999_990));
+    assert_eq!(visible.next_back(), Some(999_998));
+    drop(visible);
+    let ColumnRow::Text(last) = state
+        .column_view(Focus::After)
+        .unwrap()
+        .row(999_999)
+        .unwrap()
+    else {
+        panic!("the final added source line must remain available");
+    };
+    assert_eq!(
+        last.text,
+        Some("static const unsigned int parser_state = 1234567890;")
+    );
+    assert_eq!(last.source_row, 999_999);
+    let file = &mut state.comparison.as_mut().unwrap().files[0];
+    std::mem::swap(&mut file.before, &mut file.after);
+    file.status = FileStatus::Removed;
+    file.hunks = line_hunks(&file.before, true, &file.after, false);
+    assert_eq!(
+        (file.hunks[0].before_len, file.hunks[0].after_len),
+        (1_000_000, 0)
+    );
+    state.align_file();
+    assert_eq!(state.source_ranges[0], [(0..1_000_000, 0)]);
+    assert_eq!(
+        state.source_display_row(Focus::Before, 999_999),
+        Some(999_999)
+    );
+    assert_eq!(state.source_display_row(Focus::After, 0), None);
+}
+
+#[test]
+fn large_localized_modifications_keep_exact_hunks_and_only_diff_visible_characters() {
+    let before = vec!["unchanged generated parser body".to_owned(); 25_000];
+    let mut after = before.clone();
+    after[20_000] = "unchanged generated parser body!".into();
+    let state = live_state(before, after, true);
+    let file = &state.comparison.as_ref().unwrap().files[0];
+    assert_eq!(file.hunks.len(), 1);
+    let hunk = &file.hunks[0];
+    assert_eq!(
+        (
+            hunk.before_start,
+            hunk.before_len,
+            hunk.after_start,
+            hunk.after_len
+        ),
+        (20_000, 1, 20_000, 1)
+    );
+    assert!(state.rows.iter().all(|row| row.changes.is_empty()));
+    let display = state.after_display_row(20_000).unwrap();
+    let ColumnRow::Text(line) = state
+        .column_view(Focus::After)
+        .unwrap()
+        .row(display)
+        .unwrap()
+    else {
+        panic!("the modified source row must remain visible");
+    };
+    assert_eq!(line.changes().cloned().collect::<Vec<_>>(), [31..32]);
+}
+
+#[test]
+fn pathological_modifications_preserve_every_line_when_diff_budget_is_exhausted() {
+    let before = (0..25_000)
+        .map(|row| format!("old {row}"))
+        .collect::<Vec<_>>();
+    let after = (0..25_000)
+        .map(|row| format!("new {row}"))
+        .collect::<Vec<_>>();
+    let state = live_state(before, after, false);
+    let file = &state.comparison.as_ref().unwrap().files[0];
+    let (mut old, mut new) = (0, 0);
+    for hunk in &file.hunks {
+        assert_eq!(
+            &file.before[old..hunk.before_start],
+            &file.after[new..hunk.after_start]
+        );
+        old = hunk.before_start + hunk.before_len;
+        new = hunk.after_start + hunk.after_len;
+    }
+    assert_eq!(&file.before[old..], &file.after[new..]);
+    assert_eq!(state.rows.len(), 25_000);
+    assert!(
+        state
+            .rows
+            .iter()
+            .all(|row| row.kind == RowKind::Changed && row.changes.is_empty())
+    );
+    assert_eq!(state.after_display_row(24_999), Some(24_999));
+    let giant = "界".repeat(64 * 1024);
+    assert_eq!(
+        character_changes(&giant, "replacement"),
+        [[0..usize::MAX, 0..usize::MAX]]
+    );
+}
+
+#[test]
+fn indexed_source_ranges_match_projection_with_folds_gaps_and_unified_runs() {
+    let before = (0..60).map(|row| format!("row {row}")).collect::<Vec<_>>();
+    let mut after = before.clone();
+    after.splice(10..14, ["inserted".into()]);
+    after.splice(35..35, ["extra 1".into(), "extra 2".into()]);
+    let mut state = live_state(before, after, false);
+    state.set_context_options(2, 5);
+    for mode in [ViewMode::Split, ViewMode::Unified] {
+        state.set_view_mode(mode);
+        for focus in [Focus::Before, Focus::After] {
+            let expected = (0..state.display_rows.len())
+                .filter_map(|display| {
+                    state
+                        .display_source_row(focus, display)
+                        .map(|row| (row, display))
+                })
+                .collect::<Vec<_>>();
+            for source in 0..65 {
+                assert_eq!(
+                    state.source_display_row(focus, source),
+                    expected
+                        .iter()
+                        .find(|(row, _)| *row == source)
+                        .map(|(_, display)| *display)
+                );
+                assert_eq!(
+                    state.clamp_source_row(focus, source),
+                    expected
+                        .iter()
+                        .map(|(row, _)| *row)
+                        .min_by_key(|row| (row.abs_diff(source), *row))
+                );
+                let target = if focus == Focus::Before {
+                    Focus::After
+                } else {
+                    Focus::Before
+                };
+                let aligned = state.rows.iter().position(|row| match focus {
+                    Focus::Before => row.before == Some(source),
+                    _ => row.after == Some(source),
+                });
+                let nearest = aligned.and_then(|aligned| {
+                    (0..state.display_rows.len())
+                        .filter_map(|display| {
+                            let DisplayRow::Line(index) = state.display_rows[display] else {
+                                return None;
+                            };
+                            state
+                                .display_source_row(target, display)
+                                .map(|row| (index.abs_diff(aligned), row))
+                        })
+                        .min()
+                        .map(|(_, row)| row)
+                });
+                assert_eq!(state.corresponding_row(focus, target, source), nearest);
+            }
+            assert_eq!(
+                state.source_rows_in(focus, 9, 40).collect::<Vec<_>>(),
+                expected
+                    .iter()
+                    .map(|(row, _)| *row)
+                    .filter(|row| (9..40).contains(row))
+                    .collect::<Vec<_>>()
+            );
+        }
+    }
+}
+
+#[test]
+fn ordinary_hunks_keep_similar_anchoring_for_repeated_lines_and_eof_changes() {
+    let sources = (0..=3)
+        .flat_map(|len| {
+            (0..1 << len).map(move |bits| {
+                (0..len)
+                    .map(|bit| {
+                        if bits & (1 << bit) == 0 {
+                            "a".into()
+                        } else {
+                            "b".into()
+                        }
+                    })
+                    .collect::<Vec<String>>()
+            })
+        })
+        .collect::<Vec<_>>();
+    for before in &sources {
+        for after in &sources {
+            for before_eof in [false, true] {
+                for after_eof in [false, true] {
+                    let tokens = |lines: &[String], eof| {
+                        lines
+                            .iter()
+                            .enumerate()
+                            .map(|(index, line)| (line.clone(), index + 1 < lines.len() || eof))
+                            .collect::<Vec<_>>()
+                    };
+                    let expected = similar::capture_diff_slices(
+                        similar::Algorithm::Myers,
+                        &tokens(before, before_eof),
+                        &tokens(after, after_eof),
+                    )
+                    .into_iter()
+                    .filter_map(|op| {
+                        let (tag, old, new) = op.as_tag_tuple();
+                        (tag != similar::DiffTag::Equal).then_some((
+                            old.start,
+                            old.len(),
+                            new.start,
+                            new.len(),
+                        ))
+                    })
+                    .collect::<Vec<_>>();
+                    let actual = line_hunks(before, before_eof, after, after_eof)
+                        .iter()
+                        .map(|hunk| {
+                            (
+                                hunk.before_start,
+                                hunk.before_len,
+                                hunk.after_start,
+                                hunk.after_len,
+                            )
+                        })
+                        .collect::<Vec<_>>();
+                    assert_eq!(
+                        actual, expected,
+                        "{before:?} ({before_eof}) -> {after:?} ({after_eof})"
+                    );
+                }
+            }
+        }
+    }
+}

tests/unit/editor.rs

@@ -135,10 +135,21 @@ fn resolver_shares_native_excerpt_search_motions_and_explicit_writable_side() {
         type_text(editor, "/needle");
         editor.handle(Input::Enter);
         assert_eq!(editor.cursor().row, 70);
-        let rows = editor.diff().unwrap().source_rows(Focus::Before).count();
+        let rows = editor
+            .diff()
+            .unwrap()
+            .source_rows_in(Focus::Before, 0, usize::MAX)
+            .count();
         let cursor = editor.cursor();
         editor.handle(Input::Char(']'));
-        assert!(editor.diff().unwrap().source_rows(Focus::Before).count() > rows);
+        assert!(
+            editor
+                .diff()
+                .unwrap()
+                .source_rows_in(Focus::Before, 0, usize::MAX)
+                .count()
+                > rows
+        );
         assert_eq!(editor.cursor(), cursor);
         type_text(editor, "21g");
         assert_eq!(editor.cursor().row, 20);
@@ -4644,7 +4655,7 @@ fn editable_diff_brackets_expand_caret_excerpt_without_moving_source_anchor() {
             editor
                 .diff()
                 .unwrap()
-                .source_rows(Focus::After)
+                .source_rows_in(Focus::After, 0, usize::MAX)
                 .collect::<Vec<_>>()
         };
         assert_eq!(

tests/unit/tui.rs

@@ -4612,6 +4612,68 @@ fn hiding_diff_explorer_reclaims_workspace_and_restores_its_geometry() {
     assert_eq!(panels.files.unwrap().0, original[0].0);
 }
 
+#[test]
+fn unavailable_file_documents_show_the_actual_error_in_managed_workspaces() {
+    let mut file = deletion_file();
+    file.loaded = false;
+    let mut diff = diff_comparison(vec![file]);
+    diff.loading = false;
+    diff.error = Some("parser.c exceeds 128 MiB file limit".into());
+    let mut terminal = Terminal::new(ratatui::backend::TestBackend::new(80, 12)).unwrap();
+    let mut panels = WheelPanels {
+        workspace_managed: true,
+        ..WheelPanels::default()
+    };
+    terminal
+        .draw(|frame| {
+            panels.draw_documents(frame, &diff, frame.area(), None, None);
+        })
+        .unwrap();
+    let text = screen_text(&terminal);
+    assert!(text.contains("parser.c exceeds 128 MiB file limit"));
+    assert!(!text.contains("select the file to retry"));
+}
+
+#[test]
+fn large_generated_additions_render_first_scroll_and_final_rows_without_reparsing() {
+    let after = (0..25_000)
+        .map(|row| format!("state_row_{row:05} = 1234567890123456789012345678901234567890;"))
+        .collect::<Vec<_>>();
+    let hunks = crate::diff::line_hunks(&[], false, &after, true);
+    let mut diff = diff_comparison(vec![ChangedFile {
+        path: "parser.c".into(),
+        old_path: None,
+        status: FileStatus::Added,
+        loaded: true,
+        before: Vec::new(),
+        after,
+        before_eof: false,
+        after_eof: true,
+        hunks,
+        notice: None,
+    }]);
+    diff.files_visible = false;
+    diff.focus = Focus::After;
+    let mut terminal = Terminal::new(ratatui::backend::TestBackend::new(100, 12)).unwrap();
+    let mut panels = WheelPanels::default();
+    let mut counts = None;
+    for row in [0, 3, 12_500, 24_999] {
+        diff.reveal_source_row(Focus::After, row);
+        render_diff(&mut terminal, &diff, &mut panels);
+        assert!(screen_text(&terminal).contains(&format!("state_row_{row:05}")));
+        let syntax = panels.diff_syntax.as_ref().unwrap();
+        assert_eq!(
+            *counts.get_or_insert(syntax.after.parse_counts()),
+            syntax.after.parse_counts()
+        );
+        assert!(syntax.after.error().is_some());
+        assert_eq!(
+            diff.comparison.as_ref().unwrap().files[0].after.len(),
+            25_000
+        );
+    }
+}
+
 #[test]
 fn diff_notices_empty_comparisons_and_tiny_layouts_remain_renderable() {
     let mut terminal = Terminal::new(ratatui::backend::TestBackend::new(100, 12)).unwrap();