Repository navigation
fix(cli, docs): Refactor and enhance approximation modules and CLI support - #380
Merged
misonijnik merged 4 commits intoOct 27, 2026
Conversation
misonijnik
changed the base branch from
main
to
misonijnik/isolated-approximations-models
August 26, 2026 17:16
misonijnik
force-pushed
the
misonijnik/isolated-approximations-cli
branch
from
August 26, 2026 17:25
5b5afa3 to
7abd207
Compare
misonijnik
force-pushed
the
misonijnik/isolated-approximations-cli
branch
from
August 26, 2026 17:50
7abd207 to
5e67311
Compare
Saloed
approved these changes
Aug 26, 2026
misonijnik
force-pushed
the
misonijnik/isolated-approximations-cli
branch
2 times, most recently
from
September 7, 2026 18:41
cb6ae28 to
dbe6bd4
Compare
misonijnik
force-pushed
the
misonijnik/isolated-approximations-cli
branch
from
September 7, 2026 18:55
dbe6bd4 to
6ef9469
Compare
misonijnik
force-pushed
the
misonijnik/isolated-approximations-cli
branch
2 times, most recently
from
October 12, 2026 09:01
006bf14 to
3de3cc7
Compare
misonijnik
force-pushed
the
misonijnik/isolated-approximations-cli
branch
from
October 12, 2026 10:10
3de3cc7 to
17681bd
Compare
Gr-i-niy
approved these changes
Oct 12, 2026
Add project scaffolding, pinned dependency builds, on-demand rebuilds, compiled-output resolution, diagnostics, and documentation for custom dataflow approximations.
Three defects found by the model in formal/Opentaint/Cli/Internal/Approximation, each with the test the proof predicts. A directory counted as compiled classes as soon as a .class file existed anywhere below it. That is a test of the subtree, not a classification of the directory: a tree holding both a model project and compiled classes collapsed to one classpath root, the project was never compiled, and the scan ran without its models and reported nothing about it. Classify only when nothing below still needs building, and compute both facts in one walk rather than two (Resolve.fixed_loses_nothing, Resolve.shipped_loses_a_project, Resolve.fused_resolves_identically). A directory holding compiled classes of its own next to a project has no valid classification at all, so report it by name instead of dropping either side (Resolve.ambiguous_tree_drops_a_class). Builds staged in a directory named after the output alone, so two builders of one project shared it: one could delete the other's classes, contribute to them, and publish the mixture under a stamp recording it as up to date. Stage under a path carrying the builder's identity, and rename the previous output aside rather than deleting it, so a reader sees the old output, the new one, or nothing (Build.raceA_output_is_not_sound, Build.publishedOutputIsSound). The rebuild stamp digested the model project but not the compiler that reads it, so an upgraded CLI kept serving models compiled by the previous one. Carry the autobuilder's identity in the stamp (Stamp.shipped_stamp_serves_a_stale_output, Stamp.full_stamp_covers).
The remaining findings from the stack review. Two resolved class directories could offer the same class with different content, and nothing says which one an analysis reads, so the finding set depended on a choice the CLI never made. ResolveAll resolves every --dataflow-approximations path together, passes each directory once, and rejects a classpath whose directories disagree about a class they share. Identical copies are the ordinary result of overlapping paths and stay free (Resolve.agreement_pins_the_choice, Resolve.identical_roots_agree, Resolve.duplicate_root_is_harmless). Staging directories now live in a .opentaint-staging directory beside the output, on the skip list, so that what a killed build leaves behind is never resolved as compiled models — half-built classes applied to a scan is the same silent-loss shape as the rest of this review. The next build reclaims them, but only once the owning process is gone, so a build in progress is never disturbed; sweeping cannot reach the published output at all (Build.sweep_leaves_the_output). A build output recorded the absolute path of the project it came from, so it meant nothing on another machine. Record it relative to the output. A javac error count is a trailer for the diagnostics above it, not a diagnostic: a line that merely reads like one no longer stands in for the compiler output. Also drop a redundant sort of os.ReadDir output and a deferred close inside a loop.
Take the approximation project's pins from the batch's test project, drop rationale and CLI internals that duplicate existing instructions, and keep compilation feedback in the approximation test run.
misonijnik
force-pushed
the
misonijnik/isolated-approximations-cli
branch
from
October 27, 2026 09:57
9805d1b to
8d54627
Compare
This was referenced Oct 30, 2026
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
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.
No description provided.