testing sequence caching #54
Labels
future
Not being actively worked on, but might be worked on in the future
maintenance
Functionality-orthogonal refactoring or code cleanup that needs doing
sequence
Related to a sequence or the sequence subsystem
Milestone
Comments
|
Depends on resolution of #25. (Does Github have formal dependencies among issues/PRs? Some repository packages do...) |
gwhitney
added
the
maintenance
liammulh
added
the
future
|
This should be folded into #420. Adding milestone. |
gwhitney
added a commit
to gwhitney/frontscope
that referenced
this issue
Aug 30, 2024
In other words, it should never call calculate twice for the same index. This is tested by 10K random accesses to indices less than 1M, followed by accessing the first 10K entries, followed by accessing the last 10K entries. Hopefully that should suffice. Resolves numberscope#54.
katestange
pushed a commit
that referenced
this issue
Oct 10, 2024
* test: Implement end-to-end testing with Playwright This is an in-progress cleaned version of the final PR (#361) of the Delft student user interface project, which it supersedes. Comments from the original PR that remain relevant: * The end-to-end tests run using Firefox and Chromium. * New tests are in the e2e folder. * The tests often depend on specific classes and IDs, so they may need to be updated upon changes to Numberscope. * The tests can be executed as the following npm script: `npm run test:e2e` * An interactive testing UI and debugger can be executed as the following npm script: `npm run test:e2e:ui` Caveats concerning trying this cleaned PR and its status: * Make certain to run `npm install` after pulling this PR. * Many of the tests do not yet pass, perhaps because of the "specific classes and IDs" point mentioned above and the fact that ui2 has diverged significantly from the Delft PR series. * Tests are not yet run automatically prior to commit. * I do not think there are any image tests yet, we need to try to add them. * Tests are not yet performed in the continuous integration checks to be run on GitHub; they should be. * fix: Repair gallery/click on featured item test * fix: Repair gallery/saving a specimen test * fix: Repair scope tests. * test: Run end-to-end tests prior to any commit * test: Add image tests to e2e The strong desire for image tests led to a cascade of changes in this commit, mostly driven by the need to have reproducible images: - Removes all use of `sketch.noLoop()` and `sketch.loop()` in favor of the previously existing `stop()` and `continue()` visualizer methods, to allow: - Adds a `frames=NNN` query parameter to URLs to set the maximum number of frames a visualization may draw - Switches from the "static" instance of mathjs to a "dynamic" one, to allow its random number configuration to be controlled. In conjunction with this, moves all math functions into a single math module, as extensions of mathjs. - Removes all use of `Math.random()` in favor of the mathjs random generator - Adds a `randomSeed=AAAA` query parameter to URLs to make the mathjs random generator reproducible. - Documents all of the above changes. * doc: describe running and creating code tests Also adds new tests for `src/shared/defineFeatured.ts` and corrects the documentation extraction facility for the package manager scripts. Resolves #25. Resolves #73. Resolves #246. * test: Test that the caching mechanism won't double-calculate In other words, it should never call calculate twice for the same index. This is tested by 10K random accesses to indices less than 1M, followed by accessing the first 10K entries, followed by accessing the last 10K entries. Hopefully that should suffice. Resolves #54. * fix: Prevent ModFill from hanging on extremely large input This is an initial pass at addressing #113. Note, however, that ModFill is not reporting to the person doing visualization that it is running with different parameter values than shown. So that still must be done, but for that part we will need a resolution to #112, which will be a sufficiently involve change that we should leave it to a spearate PR from this #420. * doc: Update PR checklist. Resolves #174. * maintenance: Remove and disallow trailing whitespace in code Resolves #219. * maintenance: Make TypeScript target ES2022 in all cases. Resolves #226. * maintenance: Run typecheck in CI Resolves #292. * maintenance: Run GitHub CI on pushes to main as well Resolves #217 * test: Add OEIS Sequence-Visualizer test grid Use a list of "stressful sequences" to perform at least two image tests on each visualizer. As this change uncovered several existing errors, there are numerous other changes in this commit to return to a state of all tests passing. Here is an enumeration of other changes, in no particular order: * Change husky pre-commit actions so that the end-to-end tests are not run if there is an existing successful test since any files in the project last changed. * On most image tests, snapshot only the visualizer canvas. * Make 'axios' a runtime dependency, rather than just development, as it is used when obtaining OEIS sequence values in the running frontend. * So many other changes are made to Histogram.ts that this PR also adds hatching to show the elements with unknown factorization. Adds the "p5.brush" package to draw the hatching. Since this necessitates WebGL, adds a new 'P5GLVisualizer' base class for visualizers that wish to use a WebGL canvas. * Updates several dependencies to latest to make sure their out-of-dateness was not contributing to test issues. * Switches to the (ES) "module" import system from "commonjs" to ensure that was not contributing to test issues. Resolves #294. * Specifies font locations in a way that vite understands how to relocate in both the 'dev' and 'build' versions of the app. * Properly marks both factor cache and value cache as empty at the initialization of an OEIS sequence. * Allows ±Infinity as valid values of INTEGER param fields for convenience. * Allows specimenQuery to take the output of parseSpecimenQuery to recreate the same specimen query as was parsed, for the sake of testing. * Turns off the browser default context menu on visualizers, which wasn't particularly useful and was interfering with the UI for some visualizers. * Fixes typos in Differences visualizer (this.first -> sequence.first) * Makes FactorHistogram visualizer a p5 WebGL-based visualizer, and adds hatching to the "0 factors" bar to indicate the terms with unknown factorizations. * Uses the WebGL default controls to implement panning, zooming, and (rather uselessly but somewhat spectacularly) rotating the plot in three dimensions for the FactorHistogram visualizer. * On FactorHistogram's first pass over the sequence data, collects the counts for each possible number of factors, as opposed to the number of factors for each entry, to avoid a possibly disastrously long loop on the second pass that accumulates bin counts. This also streamlines the computation of the largest number of factors in the data. * Factors out repeated code in FactorHistogram, for labeling bars and displaying text. * Fits the hover box in FactorHistogram to the text to be displayed. * FactorHistogram displays a temporary message if factoring is taking a long time. * Moves the bar labels of FactorHistogram just under their respective bars, to make sure they are visible. * Selects y-axis ticks at round numbers for FactorHistogram, and moves the tick labels closer to their ticks. * Prevents FactorHistogram from looping except when there is mouse activity. * Prevents ModFill from freezing up if too large a "mod dimension" is chosen. Note TODO: display a warning when a smaller mod dimension than requested is actually used. * Prevents Vue's reactivity system from attempting to modify the behavior of p5 sketches, using the Vue `markRaw` method. This change prevents some instances of infinite loops caused by cascading change notifications. * Tightens up the typing of P5Visualizer so that it is possible to derive another visualizer base class P5GLVisualizer from it. Also splits up the inhabit method so that P5GLVisualizer can modify it as needed. * Allow negative start indices for Show Factors visualizer. * Show at most 100 terms in Show Factors (no room on screen for more than that). * Updates TypeScript target versions of JavaScript Resolves #226. * refactor: better typing for param descriptions * feat: Add ExtendedBigint ParamType * refactor: Remove generic parameters from top levels of Paramable hierarchy * feat: Uniform sequence bounds controls All params for controlling which terms of a sequence will be used in the visualization are removed from individual visualizers. Instead, there are uniform params in the Sequence classes themselves. Resolves #411. In this implementation, several other changes are necessary and/or expedient to make. In particular, the type of sequence indices is changed to bigint and bounds to ExtendedBigint (the union of bigint and ±Infinity), in preparation for addressing #455. Additional changes include: * Removal of SequenceDefault class, as everything now derives from Cached. * More care in caching of OEIS sequences, to provide some help with #459. * Individual-parameter validation functions now take a validation status to update based on the finding. This refactor avoids a frequent need to merge status objects. * Individual-parameter validation functions are called in a context of `this` set to the Paramable object, so that other data from the paramable can be accessed. * New `math` functions for dealing with ExtendedBigints. * NumberGlyph visualizer now tries to display as many terms as are available and will fit on the screen, except in case there are infinitely many terms available, in which case it still defaults to 64 terms. This change is needed so that the general length parameter for Sequences would affect what NumberGlyph displayed. * Puts reactivation of "known" OEIS sequences into a function rather than at top level of sequences.ts so that the timing of when it occurs can be controlled. * Renames property `_size` of P5Visualizer to `size` as multiple derived classes seem to need this value (i.e., not just used internally in the base class). * Makes sure the p5 loop is restarted if need be every time `.show()` is called on a visualizer. * Temporarily disables A000521 transversal tests until we can get incremental OEIS loading to work. * Adds a test for starting a visualization deep into a sequence. * refactor: Apply OEIS modulus on the fly, so cache can point to communal one * test: First working e2e tests with text and transversal Unfortunately, due to the intricacies of end-to-end testing with image comparison, this is a very large commit. It introduces running tests in a Docker container for the sake of reproducibility. (So note that henceforth you must have docker installed and running on your machine to perform testing, and hence to make a commit.) However, it turns out not all tests can be run in a Docker container -- Firefox is not able to supply a WebGL context inside of Docker. Hence, some tests still need to be run directly on the host machine. To deal with all this, I felt it necessary to introduce the "make" tool. In particular, creating the necessary Docker image is slow, and I didn't want to have to repeat that except when truly necessary. That, in turn, means there are now many more configuration files, as well as auxiliary files created by the Makefile. The directory tree was becoming hopelessly cluttered. To keep things straight in my head, I felt it was really necessary to reorganize the directory structure. The biggest change is to put as many of the configuration files as possible in a new etc/ directory (the old-school unix name for where to put such things). I managed to get almost everything put in there. The one major holdout is the tsconfig files; I just couldn't find a way to get TypeScript to run without its config files at the top level of the project. Slightly annoying, but then, there are so many more annoying things about TypeScript. Anyhow, accomplishing the configuration file rearrangement resulted in updating eslint. Unfortunately, it went through a _major_ rewrite from version 8 to 9, and prettier-eslint has not been updated to work with the new version. I tried to work on that update myself, but prettier-eslint is incredibly intricate in the way that it handles configurations because it is trying to provide a great deal of flexibility and power, and it is trying to infer prettier configuration from eslint configuration (and maybe even vice-versa as well, I'm not sure). Since we were only using a tiny fraction of that power, it turned out to be significantly easier to just replace prettier-eslint with a custom script tools/prettiest.js that runs first prettier, then eslint. It also meant we could get rid of the lint-staged tool, so there is that. (While I was at it, I ran the "depcheck" tool and got rid of some other unneeded cruft that had accumulated.) But now it all works, and as a side benefit since just about every npm script now runs through make, you should never need to worry about running install before dev or build before preview, etc. Make keeps track of what depends on what and whether/how to redo the prerequisites before taking the requested action. Even better, if you just ran successful end-to-end tests, which take a while now, then `git commit` will not have to re-run them. Those are the major points. In addition, there are a number of minor changes resulting from errors uncovered in getting all the tests to work, etc: - Reordering of element attributes and other reformatting in vue components, because with the lint updates and reconfiguration, it now seems significantly stricter about vue templates. - First pass at an updated User Guide for ui2; it definitely needs more work. - Some other more minor documentation updates. - Updates tools/editor/autoformat.el so that emacs can use the new "prettiest" tool for formatting (even though I know I am likely the only one in the world who will ever use that). * test: Add skipped tests reflecting known shortcomings As per Numberscope discussion today, this is the last missing element of the end-to-end testing PR. With these skipped tests, we are flagging that there are known concerns we want to resolve at least by beta, if not alpha. Also fixes a small gap in the docs for running from source that Aaron noticed in the meeting. With this, the PR will be marked ready for final review. * fix: Generate docker image even with no existing test results * chore: Attempt to run e2e tests in CI workflow * test: No WebGL tests in CI :-( * fix: format of Playwright reporter parameter * chore: Try to grab GitHub CI actuals to make separate CI test file * fix: oops test must succeed to generate artifact * chore: Another try to grab ci snapshots * test: Add in the extracted CI snapshots * doc: Note that PR reviewers must run e2e tests themselves, too. * chore: Stop grabbing snapshots when we don't need them * fix: Changes as per review comments * doc: more info on Docker (and fix remaining typo) * test: Fuzz the pixel comparison in Firefox WebGL tests * test: Really fuzz the pixel comparison in Firefox WebGL tests * test: Check if docker tests passed before updating result directory * doc: Improvements per code review * doc: show the expected output of successful end-to-end test * chore: remove stray comment that no longer applies * fix: useful default camera controls for WebGL: left drag pans, wheel zooms * fix: correct Husky action inclusion and test it * fix: don't alter the URL just loaded, and reset frame limit on changes * fix: adjust dragging, detected mouse position, and text size for zoom/pan * fix: Keep the 'too many bins' message in a fixed absolute canvas position
|
Completed in ui2 thanks to #420. Closing. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Once the test framework is up and running, there should be a test that the caching of sequences (SequenceCached.ts) never calls calculate() twice for the same index n. The random integer sequence functionality (PR #53) relies on this guarantee, since we wish to guarantee that the randomly generated value is the same if the same term is called twice. But its calculate() function is just a call to Random(), so without careful caching, this guarantee would not be met.
The text was updated successfully, but these errors were encountered: