side_diff: fix the column layout overflow and --expand-tabs - #270
Open
sami-daniel wants to merge 8 commits into
Open
side_diff: fix the column layout overflow and --expand-tabs#270sami-daniel wants to merge 8 commits into
--expand-tabs#270sami-daniel wants to merge 8 commits into
Conversation
The side-by-side layout tests need parameterized cases so that each width and tab size shows up as its own named test instead of a loop that reports a single failure with no indication of which input broke.
`Config::new` sized the side-by-side columns with signed arithmetic on the `--tabsize` value, which `params.rs` accepts with no upper bound. `tab_size + GUTTER_WIDTH_MIN` overflowed for a large one: with `-C overflow-checks=on` that aborted, and in release it wrapped and drew bogus widths. A value above `isize::MAX` did not even abort, it cast to a negative number and silently laid the columns out as if the tab size were small. Closes uutils#264. The calculation moves into `Config::layout`, in `usize` throughout, and holds for every input on its own rather than relying on validation upstream: - An early return for `tab_size > full_width`. This is exact, not a clamp: the offset is a multiple of `tab_size`, so it is either zero or already past the right edge, and neither leaves room for a second column. It is also what keeps the sum below in range. - `saturating_add` for the gutter sum, which covers the one case the strict `>` lets through, `full_width == tab_size == usize::MAX`. - `saturating_sub` for the two bounds that could go negative. `full_width = 5, tab_size = 8` really does produce an offset past the right edge. - `usize::midpoint` for the balance point, which cannot overflow. - `saturating_sub` on `separator_pos`, which underflowed for a zero width. Unreachable from the CLI, reachable through the library. A tab stop every zero columns has no meaning and every reader of `tab_size` divides by it, so `Config::new` now normalizes it to one. The layout alone was not enough: fixing it only moved the division by zero from `layout` to `format_tabs_and_spaces`. The new `mod layout` covers the boundaries, including the two that pin the design down: `full_width = 10, tab_size = 8` still leaves two columns, so no looser guard is correct, and `tab_size == full_width` yields a non-zero offset with an empty half line.
`Config::new` took `expanded` but ignored it when sizing the columns. The manual's "Preserving Tab Stop Alignment" section explains why it matters: column two has to start on a tab stop only so that tabs in the right column keep their position relative to the stops. With `--expand-tabs` there are no tabs left in the output, so there is no grid to preserve and every column is a stop. Compared against the output of `diff` from GNU diffutils 3.10. The separator column now matches it for `-y -t` at widths 40, 80 and 130, and for `-y -t --tabsize=4 --width=100`: columns 19, 39, 64 and 49. At the default width it sat on column 62 before, two columns off. The tab size still decides how far an expanded tab reaches, so it stays in the field and only `layout` sees the 1. Feeding the 1 into the field would shrink every expanded tab to a single space. The widened half line makes a crash reachable. A tab size near `usize::MAX` used to yield an empty half line, so `process_half_line` returned before drawing anything; with `-t` the line now fits and three expressions of the form `current_width + tab_size - (current_width % tab_size)` overflowed. `format_tabs_and_spaces` and the tab arm now measure the step to the next stop against the room that is left instead of summing absolute columns, which the surrounding `current_width <= max_width` already bounds. `test_full_width_40_tab_8` used `expanded = true` and expected the widths computed while ignoring it. The separator lands on column 19 either way, so only the half width and the offset change. The new tests were checked by mutation. Dropping the ternary breaks `expanded_tabs_widen_the_half_line`, `expanded_tabs_lay_out_as_a_stop_on_every_column` and `test_full_width_40_tab_8`; applying it to the field instead breaks `expanded_tabs_keep_the_real_tab_size_for_rendering` and `expanded_tabs_reach_the_next_real_tab_stop`.
`width` and `tabsize` were commented out of `fuzz_side`, along with a `width == 0 || tabsize == 0` early return that was commented out too, because the column arithmetic could not take arbitrary values. It can now, so both are fed to the target and neither guard is needed. They are `u16` rather than `usize`: the layout handles the whole range, but a width near `usize::MAX` asks the renderer for petabytes of padding, which would only produce timeouts.
sami-daniel
changed the title
--expand-tabs
The assertions state the widths already, and the surrounding test name carries the tab size.
A carriage return pads all the way to column two, the only caller that reaches the far end of the line. That walk used to add absolute columns and overflowed with `--width` and `--tabsize` at the top of the range, which aborts under `-C overflow-checks=on`. Both values sit at the maximum on purpose: a smaller tab size walks to that end one stop at a time, and a smaller width never reaches the sum that overflowed. Expansion stays off, since it pads with spaces one column at a time and would not finish at this width.
sami-daniel
marked this pull request as ready for review
August 13, 2026 03:43
|
GNU diffutils testsuite comparison: |
A tab size wider than the line collapses the layout to a single column and draws nothing, so the whole `usize` range costs no more than a small one. Capping it at `u16` left out the values above `isize::MAX`, which are the ones the old signed arithmetic turned negative. The width stays a `u16`, since the padding it asks for is written one column at a time and the cost scales with it.
Superscript digits do not survive every terminal or editor the file gets read in.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes #264.
Side-by-side sized its columns with signed arithmetic on the
--tabsizevalue, which is parsed with no upper bound. A large tab size overflowed, and what happened next depended on the build: with-C overflow-checks=onthe process aborted, and in release the value wrapped and the columns came out with bogus widths.A tab size past the signed range never aborted. It cast to a negative number and sized the columns as if the tab size were small, so the output looked ordinary and was wrong. The calculation is now unsigned throughout and holds for every input on its own, instead of relying on validation upstream.
--expand-tabswas accepted but not taken into account when sizing the columns. Column two has to start on a tab stop only so that tabs in the right column keep their position relative to the stops, and once tabs are expanded there are none left to keep aligned.This changes the output of
-y -t. At the default width the separator moves from column 62 to 64.That wider half line exposed a third overflow, introduced and fixed in the same commit. A large tab size used to leave the half line empty, so nothing was ever drawn; with
-tthe line now fits, and the sums that walk to the next tab stop overflowed. They measure the step against the room that is left instead of adding absolute columns.