Repository navigation
Clear the issue backlog (4.3.12) - #1235
Merged
Merged
Conversation
A run counts as failed when the engine exits with an error or err.log holds a located error -- and the engine has usually written the trees of every pass it got through. The failure branch refreshed only the output and log views, so the sequence view's per-pass tree icons kept whatever state the previous run left: passes that did produce trees offered none on a first run, and the output view listed no trees. Now a failed run refreshes the sequence and text views, switches the output view to its tree listing when any trees exist, and logs how far they go. Checked with a pass that hits a run-time error: 14 trees written, run reported failed. Closes VisualText#785. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
"Generate Rule General/Exact" worked from text (.txt), highlighted text (.txxt) and pass files, but not from the place a rule is usually designed: a pass's parse tree. findSelectedTreeStr had no .tree branch, so the command did nothing there, and the editor menu did not offer it. In a .tree file the selected lines are parsed directly (parseTreeLines, as Generate Path does) and the top-level nodes become the rule's elements. The rule goes into the pass the tree belongs to (anaNNN.tree -> pass NNN, as Generate Path does); final.tree falls back to the current pass. Both commands are now in the .tree editor menu. Closes VisualText#475. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…lText#328) Going through every setting the extension declares and every one it reads turned up two defects and a set of descriptions that did not say what the setting does: - analyzer.runMode (the status-bar run mode: interpreted / compiled / compiled-analyzer / compiled-kb) was read and written but never declared. VS Code refuses to write an unregistered setting, so the mode was lost on every reload; the refusal is an async rejection the surrounding try/catch could not catch. Declared as an enum, the write's rejection is now handled, and the integration test that guards written settings covers it. - createtmpfile.tmpDir, another extension's setting copied in with a snippet, chose where partial trees were written. They go to the system temp folder, which is what it fell back to. - update.auto, textView.fast, analyzer.current, engine.visualtext, engine.analyzers, user.name and logs.clear get descriptions that say what they do and which ones the extension sets itself. Every declared setting is read somewhere. Keys are not renamed: moving the un-prefixed ones (update.*, user.name, ...) under nlp.* would silently reset everyone's values. Closes VisualText#328. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…alText#613) "Sort & Unique text lines" sorted and de-duplicated but threw the counts away, and nothing counted words across a corpus. - Sort & Count text lines (editor menu, beside Sort & Unique): each distinct line with how often it occurs, most frequent first, as "count<TAB>line" in a new editor -- the text itself is untouched. Works on the selection or the whole file. - Count Words in Folder (Text view, on a folder): word frequencies over every file under it, skipping the engine's _log folders, with a header giving the total, distinct and file counts. A word is a run of letters and digits in any script, inner apostrophes and hyphens kept, lowercased. Counting is pure (src/output/counts.ts) and unit-tested; an integration test runs Sort & Count in VS Code. Closes VisualText#613. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…isualText#658, VisualText#794) VisualText#658: the output view could copy a file only to another analyzer's kb/ or input/ folder. "Copy or Move File To..." (output item context menu) asks where with a save dialog, starting in the analyzer's input folder, then whether to copy or move. VisualText#794: a pass file a run produced -- a generated rule file, a RUG pass -- had to be found again through Select Library Pass to get it into the sequence. Output .nlp/.pat files now carry "Add to Analyzer Sequence": pick the pass to insert after; if a pass of that name is already in the sequence, either replace it (it keeps its place, only its file changes) or add this one under the next free name (name2, name3, ...). Closes VisualText#658. Closes VisualText#794. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The output view showed one input file's outputs at a time. After analyzing a folder, each file's _log folder had to be reached by selecting the files one by one. Selecting a folder in the Text view now lists every input file under it that has outputs, each expanding to its own outputs in the current mode (all / matches / trees). Selecting a file goes back to that file's flat list. KB and orphan-pass modes are per analyzer and unchanged. The file actions (delete, rename, copy, add test) are not offered on the input-file rows, so they cannot act on the input itself. Also fixed on the way: - The matches and trees modes reused the _log folder the "all" mode had last found, and listed nothing if "all" had not run first. Each mode now finds the folder itself. The trees mode no longer lists final.tree twice. - fileHasTest set the log folder instead of the test folder, so it always answered false. Closes VisualText#688. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
) At activation the updater checks, downloads and unpacks the engine, ICU libraries, VisualText files and example analyzers. It reported only to the log view, so a first start that downloaded the engine looked like an extension that did nothing for a minute. While the updater runs, the status bar now shows a spinning "NLP++" item naming the step in progress -- "Downloading the NLP++ engine", "Unpacking the example analyzers" -- and it goes away when the queue finishes or the user stops it. Status-bar progress rather than a notification, so an ordinary start that finds nothing to do is not interrupted. Closes VisualText#230. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
) The engine is downloaded into the extension's own folder, ~/.vscode/extensions/dehilster.nlp-<version>/nlp-engine. Which <version> was decided in three places: at activation the newest version found on disk; then, only if auto-update was on, the running extension's version; and a lexicographic sort in getLatestExtPath that also wrote it. With several versions on disk -- after an update VS Code has not cleaned up yet, or a downgrade, or with update.auto off -- a new version could run an older version's engine, or download into another version's folder. Removing the older versions was the workaround the issue describes. (This machine had seven.) The version is now the running extension's, set once at activation and never replaced by a directory scan. getExtensionPath() (settings init, the example-analyzers path, engine updates) prefers the running version's folder, falling back to the newest installed and then to the code's own location as before. getLatestExtPath is gone. The folder layout is unchanged, so F5 debugging still uses the installed engine of the same version. Closes VisualText#481. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…sualText#295) VisualText#295 reported wrong colorization in the partial tree ("View selected tree"), with no repro. Checked against today's tree grammar: 1,230 partial trees cut from 15 real final.tree files exactly as findTreeFileLines cuts them, 30,469 lines, tokenized with vscode-textmate beside the full trees they came from -- every line gets the same scopes in both. The grammar colors a tree line by line, so a partial tree cannot color differently any more. What the check did find is in the cut itself. findTreeFileLines split each line on its first "[" to read the offsets, and the engine writes a literal bracket token as " [ [2,2,2,2,0,0,punct]": the token was taken for the field list and the line dropped, so the partial tree of "x [y] z" lost its "[". The cut now goes through the shared, tested parseTreeLine (src/trace/treeParse.ts), which already knew that case, via a new treeLinesInRange -- tested on the engine's own output for that text. Closes VisualText#295. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
VisualText#98 asked for a manifest that bundles .nlp passes and KB files. That is the .nlm mod file: sections under "<modfile kb\user\x.dict>" headers, unpacked by the extension into kb/user, spec and input, with the passes inserted into the sequence after a pass the user picks (Mod Load / Create / Add in the KB, sequence and text views). The engine needs no change -- the extension does the unpacking. But the format only worked on Windows. Headers use backslashes, and sections were classified by looking for "\kb\user\", "\spec\" and "\input\" in a joined path. On macOS and Linux nothing matched, so every section was "unknown" and its pass never went into the sequence, and path.join kept the backslashes, writing a file literally named "kb\user\en-numbers.dict" in the analyzer folder. Building a mod file there recorded the path with "/". Paths are now split on either separator (src/session/modPaths.ts, tested), and headers are always written with backslashes so a mod file made anywhere loads in older versions too. Checked against the shipped languages/English/en-numbers.nlm: kb, input and spec sections all recognized. Closes VisualText#98. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Version bump and CHANGELOG for the twelve issue fixes on this branch. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…sualText#272) A pass that builds a knowledge base -- SaveKB("mykb.kbb", ...), or a .dict written with << -- puts the file in the run's <text>_log folder, where the analyzer's kb/user never sees it; using it meant copying it by hand and running again. - "Toggle KB Generator" on a pass in the sequence marks it (shown as "KB generator"); the marks live in the analyzer's state.json. - With nlp.analyzer.regenerateKB on (off by default), after a successful run in which a marked pass has changed since its output was last copied, the .kbb/.dict files named in that pass's own text are copied from the log folder into kb/user and the text is queued once more. The second run finds nothing changed, so it does not repeat. The re-run waits for the current queue, since a second queue entry for the same file would share the first one's status. Checked with the engine: a pass writing words.dict -> run 1 leaves the attribute empty; with words.dict copied into kb/user, run 2 reads it (animal=1). File-name extraction and change detection are pure and unit-tested (src/session/kbGen.ts). Closes VisualText#272. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
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.
All thirteen open issues, one commit each, released together as 4.3.12 so the version bumps and the CHANGELOG don't collide. Each commit message gives the details and how it was checked.
.treefile; the rule goes into that tree's passanalyzer.runModewas never declared, so the run mode was lost on every reload; straycreatetmpfile.tmpDirsetting removed; descriptions fixed. Keys are not renamed, because renaming would reset everyone's values[tokens dropped from partial trees.nlmmod files; they were Windows-only and now work on macOS/Linuxnlp.analyzer.regenerateKBon (default off), the .kbb/.dict files a changed generator pass writes are copied from the log folder into kb/user and the text runs once more. Premise checked with the engine (run 2 reads the dict run 1 wrote)Help/markdown/writekb.mdon visualtext-files main, under Knowledge Base FunctionsTested:
npm test(all suites; new unit tests for counts, tree ranges, mod paths, last tree pass), integration 57/57 (new: Sort & Count, run-mode registration), andvsce-packagewith the icon check.Not exercised in VS Code: the dialog/quick-pick flows (#658, #794), the folder browser (#688) and the updater progress (#230), and the KB generator re-run (#272), which needs an installed engine for the running version. All are typechecked and linted, but the harness can't click through them.
Closes #475
Closes #794
Closes #658
Closes #688
Closes #613
Closes #785
Closes #230
Closes #481
Closes #328
Closes #295
Closes #98
Closes #626
Closes #272
🤖 Generated with Claude Code