Repository navigation
chore: merge upstream/main and release v4.5.0 - #113
Merged
Merged
Conversation
Rust (src-tauri/src/): - blob_capture.rs: memory scanner thread, blob processing, state sync - reward_watcher.rs: EE.log tail, OCR trigger, relic reward detection - monitor.rs: monitor catalog, inventory state, crafting jobs - catalogue.rs: item catalog, blueprint names, weapon catalog - companion_api.rs: Warframe companion API integration - diagnostics.rs: debug tools, memory probe, overlay diagnostics - log_watcher.rs: EE.log parsing, trade detection, riven events - rivens.rs: riven analysis, OCR, memory watcher - settings.rs: settings load/save, window state management - wfm_commands.rs: warframe.market API commands - wfm_queue.rs: price fetch queue with priority support - worldstate.rs: worldstate alerts, syndicate alerts - Additional: credentials, db, image_cache, inventory_state, paths, platform, pricing, refresh, relic_pick, stats, syndicates, trade_log, updater, wfcd, wfm, wfm_top TypeScript (src/): - hooks/: useCatalog, useInventoryData, useMarketData, useSettings, useOverlays, useModularWindow, useFissureNotifications, useTimerPreferences - constants/: categories, tauri - shared/: ResizeHandle component - lib/: rivenWindow helper - Refactored App.tsx, MarketHelper, Reports, Foundry, and others lib.rs reduced from ~10k to ~700 lines.
Centralize all Windows-specific code into platform/ and ocr/ modules with trait-based abstractions for Linux portability. Platform module (platform/mod.rs, windows.rs, linux.rs): - CredentialStore trait: save/load/delete credentials - ProcessAccess trait: find PID, open process handle - ProcessHandle trait: query regions, read memory, enumerate regions - WindowManager trait: find HWND, get window rect, set overlay topmost - LocaleProvider trait: get system locale - ComInit trait: initialize COM for WinRT/OCR - FileWatcher trait: watch files for changes - MemoryRegionInfo struct with is_committed, is_readable, is_writable, is_executable, is_image fields OCR module (ocr/mod.rs, windows.rs, linux.rs): - Business logic in mod.rs: preprocessing, BMP conversion, matching, scoring - Platform-specific capture and OCR engine in windows.rs - Linux stubs in linux.rs - Removed circular dependency between platform and ocr Consumer refactoring: - credentials.rs: Platform::CredentialStore - diagnostics.rs: Platform::WindowManager, ProcessHandle - settings.rs: Platform::CredentialStore - memory_scanner.rs: Platform::ProcessAccess, ProcessHandle, RegionSource - monitor.rs: Platform::ProcessAccess, ComInit - rivens.rs: Platform::ProcessHandle - companion_api.rs: Platform::ProcessHandle - log_watcher.rs: ocr:: capture functions - mem_regions.rs: open_region_source factory, RegionSource::stats() Result: 52+ #[cfg] blocks consolidated, all 4 OCR tests pass.
Upstream splits its `lib.rs` into domain modules, which the fork had already done along different boundaries. The fork's split stands; what the merge takes is the new platform abstraction. `platform.rs` becomes `platform/`, with the Linux implementations wired to the code that already existed: the `/proc` PID lookup, the Secret Service store and the X11 window rect. `CredentialStore` needs a sync entry point, so `credentials` grows one and its async commands go through the trait. Game-running status now follows the PID instead of waiting for a blob parse to succeed, which is upstream's fix for the status reading "not running" while the game was already up. Upstream's `blob_capture` and `reward_watcher` modules are left out; `monitor` holds both loops here. `ocr.rs` stays one file rather than adopting upstream's `ocr/` split.
- Fix read_at resume address bug (short reads no longer skip gaps) - Replace enumerate_regions_from Vec with regions_from iterator - Add read_into borrowing method to ProcessHandle trait - Add max_len parameter to RegionSource::read_at - Route WindowsRegionSource through Platform::open_process - Replace is_image bool with RegionBacking enum - Change open_process to return Result with error details - Exclude volatile sections (RecentVendorPurchases) from baseline rejection - Add content hash deduplication for unchanged blobs
- ProcessRegions: chunked reading for large regions (> read_cap) - ProcessRegions: deadline parameter for timed walks - ProcessRegions: caller-supplied filter for region selection - RegionSource: region_at() default method to check if addr is mapped - RegionSource: read_at contract documented (bytes must live at addr) - RegionBacking::Kernel: per-platform semantics documented - MEM_MAPPED intentionally maps to Anonymous (not File) - CredentialStore: blocking warning doc for locked keyring - try_cached_blob: use region_at for fast unmapped-address detection
`ProcessHandle` now lends into a caller-owned buffer through `read_into`, enumerates mappings through `regions_from`, and `open_process` reports why it failed rather than returning `None`. Move the Linux process and mapping code out of `memory_scanner_linux` into `platform::linux` behind that trait, mapping `/proc/pid/maps` entries onto `MemoryRegionInfo` and `RegionBacking`. The scanner keeps its own `RegionSource`: `ProcessRegions` reads only the first `read_cap` bytes of a region, which loses a blob sitting past that offset in a large `[heap]` mapping.
`ProcessRegions` now chunks large regions, takes a region filter and a deadline, and can answer `region_at`, so the Linux scanner drops its own region source and drives the shared one. The two-tier walk opens one `ProcessRegions` per tier instead of partitioning a mapping snapshot, and the cached-blob probe wraps the source in `StitchGuard`, which stops the stitch at a mapping whose bytes must not be spliced into the blob. `read_at` reads blind, so without the guard the stitch runs out of the heap into code or a mapped file. Tests fake the process handle rather than the region source, which keeps the chunking, the minimum region size and the tier filters under test.
Splits the ~10k-line lib.rs into focused domain modules (app_state, blob_capture, catalogue, monitor, diagnostics, settings, wfm_commands, worldstate, and more) and introduces a platform abstraction layer (ProcessAccess, ProcessHandle, WindowManager, CredentialStore, LocaleProvider, ComInit, FileWatcher) so Windows-specific code funnels through platform::windows instead of being scattered across the backend. OCR is split the same way into ocr/windows.rs and ocr/linux.rs behind ocr/mod.rs. Preserves the v4.4 section-baseline blob validation and background catalogue refresh behavior. Includes fixes from two rounds of review by Lyrex (Linux fork maintainer), who exercised the new traits against a real second implementation rather than reading the diff in isolation: - read_at resume-address bug: a short/faulted read returned the region end as the next address, letting the stitcher splice non-adjacent memory into one blob - region enumeration switched from an eager Vec to a lazy iterator (regions_from), restoring the early-exit behavior of the pre-refactor per-region loops (notably the 200ms trigger-scan loop) - read_memory replaced with a borrowing read_into to drop per-read allocation in hot loops - read_at gained a max_len bound tied to the caller's remaining budget - WindowsRegionSource now opens its process handle through Platform::open_process instead of calling OpenProcess directly - is_image: bool replaced with a RegionBacking enum (Anonymous / File / Kernel) with documented per-platform semantics - open_process returns Result<_, String> with GetLastError() detail instead of discarding the failure reason - read_at's contract now documents that returned bytes must live at the requested address, and added a region_at() method so callers can detect unmapped holes instead of skipping over them - regions larger than read_cap are now read in chunks across multiple next_region() calls instead of silently losing everything past the cap (was breaking on large single-mapping heaps) - documented that CredentialStore implementations may block on user interaction (locked keyring prompts) PR #75 (branch feat/backend-refactor, Oost-hash), squash-merged locally. Not pushed to GitHub. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
WFCD's warframe-items repo stopped putting a `name` field on components[] entries (only uniqueName + itemCount remain, verified live against Warframes.json/Primary.json/Pets.json). This broke three call sites in wfcd.rs that expected it: - Pass 2 component flattening skipped every single component via `continue`, so Chassis/Systems/Neuroptics/Barrel/etc. never made it into `items`/`display_names` at all. - The top-level recipe tree builder and build_recipe_node's WFCD fallback branch then had nothing to fall back to but the literal string "Unknown", which is what showed up in the Foundry ingredient list (affects both v4.4.0 and v4.4.1 — this predates the PR #75 refactor, it's an upstream data change). Fix: derive the missing name from the component's own uniqueName path relative to its parent's (".../AshPrimeChassisComponent" under parent ".../AshPrime" -> "Chassis" -> "Ash Prime Chassis"), used only when WFCD's own `name` field is absent so a future reversal upstream is still honored. Verified against live Ash Prime, Braton Prime and a Kubrow/Kavat part — output matches the historical naming convention exactly. cargo test --lib: 76/77 (same pre-existing unrelated paths.rs failure as before this change). pnpm build clean. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Following up on e0ab70f: drop the middle fallback that trusted components[]'s raw `name` string when the catalog lookup missed. WFCD no longer sends that field at all (confirmed: 0 components anywhere carry it), and even if it comes back it shouldn't outrank resolving the component's own uniqueName against the catalog — that's the canonical identifier either way. Now it's strictly: catalog lookup by uniqueName first, then path-derivation, nothing else. cargo test --lib: 76/77 (same pre-existing unrelated paths.rs failure). cargo check clean. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
…e most complete one
Multiple copies of the FULL_ACCOUNT blob can be readable in memory at
once (e.g. a buffer left over from before a re-serialize, still fully
intact and not yet overwritten, alongside the live one).
blob_seed_offsets()'s `{"` check only catches a stale copy whose
opening brace was already clobbered — it says nothing about a stale
copy that's merely older but still completely valid JSON.
capture_all_blobs/stitch_blobs previously committed to whichever
candidate scan finished stitching first and immediately dropped every
other in-flight scan once that happened, including ones opened before
it. That's a pure timing race with no check that the winner is
actually the current state — a still-intact stale snapshot can win
over the live blob just by having its end marker turn up sooner,
silently rolling the displayed inventory back ("items dropped" after a
scan, reported recurring on 4.4.1).
Fix: outside diagnostic `save` mode, every scan already in flight is
left to resolve (succeed or get dropped for growing too large) instead
of being killed off the moment another one completes, and new seeds
keep being opened for a bounded window after the first candidate
resolves (MAX_CANDIDATE_SEEDS / EXTRA_SEED_BUDGET) instead of stopping
dead. Once nothing is left in flight, the candidate with the most
total items/mods is sent — not whichever happened to finish first.
`save` mode (blob debug dumps) is unchanged.
Added two regression tests exercising both scan orderings; verified
the lower-address-first one reproduces the bug (credits: 1 instead of
2) against the pre-fix logic before restoring the fix.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Includes the stitch_blobs race-condition fix (843c437) for the recurring "items dropped after scan" bug reported on 4.4.1. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
…trip component_suffix_from_path assumed a component's own path tail was always literally prefixed by its parent's path tail. That's false whenever DE's path segments use a different word order than the parent's own tail (e.g. parent tail "PrimeAkstiletto" vs. component tail "AkstilettoPrimeBarrel"), producing garbled catalog names like "Akstiletto Prime AkstilettoPrimeBarrel" instead of "Akstiletto Prime Barrel". Confirmed via live WFCD data for Akstiletto Prime, Vectis Prime, and Aksomati Prime. Fix: split both the component's and parent's path tails (plus the parent's display name) into words and subtract the parent's words from the component's, order-independently, instead of a literal prefix-strip. Falls back to the untouched (now word-split) tail when nothing lines up, same as before. component_suffix_from_path/component_name_from_path now take the parent's display name so they can subtract on it too. camel_to_words in catalogue.rs made pub(crate) so wfcd.rs can reuse the same PascalCase word-splitter. Added 3 unit tests: reversed word order, the original matching-prefix case, and the no-overlap fallback. cargo test --lib: 82/82 except the one pre-existing unrelated paths.rs failure. pnpm build clean. This was found while investigating a user report of relic reward icons showing a grey "?" placeholder. Static/cache analysis of the current (already-fixed, 4.4.2) catalog data shows the specific flagged item (Akstiletto Prime Barrel) actually resolves to a valid image via the parent-weapon-image fallback already in Pass 2 — so the garbled name alone likely wasn't the direct cause of that screenshot (a stale cache/build predating e0ab70f/4c7330e is the more likely explanation). This fix is still worth having independently: it removes a fragile dependency on fuzzy substring matching accidentally still finding these components by their garbled name, and fixes the display name shown anywhere it surfaces (Foundry lists, search, tooltips). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
…hema change WFCD's components[] sub-objects stopped carrying ducats (same upstream change that dropped name/imageName), but unlike those two fields there was no fallback: comp_ducats was pushed as None unconditionally, with no parent-item value to fall back to (ducats aren't stored on the parent item either). This silently broke the item.ducats.is_some() tradeable-item heuristic used in wfm_queue.rs (gates the entire background WFM price-fetch queue) and wfm_commands.rs (gates whether a fetched price persists to disk), plus the ducat-value maps in monitor.rs and relic_pick.rs (ducat_score is the default sort order for the relic-pick overlay's relic list, not just the opt-in "ducat" best-pick priority). Fix: ducat value is a pure function of drop rarity (Common=15, Uncommon=45, Rare=100, mixed=25/65) — the Warframe Wiki computes it the same way rather than storing it independently, and we already have the rarity data from relic_rewards. Backfill it locally from data already fetched, no live wiki dependency added. A small ~25-entry bundled exceptions table (resources/ducat_exceptions.json, fetched once from the wiki's Module:Void/data DUCAT_EXCEPTIONS table) covers the handful of parts whose value doesn't follow the formula. Runs after relic_rewards is built and before blueprint_names reads item.ducats, so both components and blueprint entries get backfilled in one pass. wfm_queue.rs/wfm_commands.rs/monitor.rs/relic_pick.rs all read the same backfilled wfcd_items list, so no changes needed there. cargo test --lib: 85/86 (pre-existing unrelated paths.rs failure). pnpm build clean. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Includes the ducat-value backfill fix (1442b1c). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
`ProcessRegions` re-read `/proc/pid/maps` for every region it visited, which stretched a Linux walk of a Proton game to seconds per tier. A failed chunk read also dropped the rest of its region, so a fault early in a large mapping hid any blob further in.
- Resolve quantities/mods authority contradiction between memory scanner and companion API gap-fill (companion API is fully suspended, so this is build-correctness only, not a live behavior change) - Merge the two independent EE.log tailers into one read loop - Centralize Tauri event names in events.rs, drop confirmed-dead events - Add a shared read-modify-write persistence helper and migrate inventory_state/companion_api/catalogue writers onto it - Bump version to 4.4.6 across package.json, Cargo.toml, tauri.conf.json - Add CONTRIBUTING.md documenting the dev -> main branch workflow now that main is protected by a ruleset (PR + 1 approval required) Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
MiscAccountData can legitimately disappear from the FULL_ACCOUNT blob between scans (same class of behavior as RecentVendorPurchases), so it was being wrongly treated as a truncated-capture signal. Add it to VOLATILE_SECTIONS so its absence no longer triggers a rejection. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
… bump to 4.4.8 Component names were derived from uniqueName path tails after WFCD dropped components[].name; DE's path token for Neuroptics is "Helmet". Names now come from ExportResources.json + dict.en.json, ExportRecipes resultType, and Relics.json reward items. Path derivation is removed; unresolved components are skipped and logged. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Trade detection (log_watcher.rs): - Restore the single-stage `received_raw` extraction in parse_trade_dialog. The Sep 22 backend-modularization refactor (eaaaa42) turned it into a two-stage extraction that re-searched the already-stripped substring for "the following:" — a string that could never be found there again by construction. This silently zeroed every trade's received side: plat purchases logged nothing at all, sales lost their received platinum and were misclassified as generic "trade". - Strip trailing private-use-area glyphs (e.g. U+E000 rank/status icons) from the parsed player name so whisper auto-matching isn't broken by an invisible character. - Added regression tests using real (redacted) EE.log dialog text from a user report, covering both the sale and purchase cases. Item icons (wfcd.rs): - Parts and blueprints (e.g. "Neuroptics", "Prime Neuroptics Blueprint") were showing their parent item's icon because WFCD's components[] lost imageName in an upstream schema change. Added an ExportResources-based icon lookup by uniqueName (NameResolver.icons/icon()), falling back to the blueprint's ExportRecipes resultType, and use it to override Parts/Blueprints images. ~1200 items get their correct icon back. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Closes the second half of the trade-logging bug fixed in f709769. The clippy `too_many_arguments` refactor in d0a6c74 grouped several Tauri command args (add_trade, wfm_create_riven_auction, wfm_receive_tokens) into structs, which changed the expected IPC shape, but the frontend invoke() calls were never updated to send { params: {...} } and the failures were silently swallowed by an empty .catch(). Trades parsed correctly from EE.log but never reached the database. - Add #[serde(rename_all = "camelCase")] to AddTradeParams, RivenAuctionParams, ReceiveTokensParams - Wrap frontend invoke() calls in { params: args } and log failures instead of swallowing them - Emit a trades-updated event so the Reports tab refreshes live Also bundles two disclosed cleanup commits from the same PR: a clippy warning pass (24 warnings, mostly mechanical + parameter-grouping structs, no behavior change) and a build-warning pass (static vs. dynamic imports in 3 frontend files). Squash-merged from PR #77 (Oost-hash), authored by the original contributor.
fix: preserve trade dialogs across log reads
WFCD dropped name, imageName and ducats from components[], so the catalogue skipped every component on live data. Components now take their names and icons from DE's ExportResources and dict.en tables, and their ducats from relic drop rarity plus a bundled exceptions table. add_trade and wfm_create_riven_auction now receive their arguments wrapped in params with camelCase keys. The frontend used to send them flat, Tauri rejected the call and the error was swallowed, so in-game trades never reached the database. Upstream's single EE.log tailer and its reward_watcher, blob_capture and console login modules do not fit the fork's layout and stay out. The frontend still starts the fork's log watcher. The section baseline ignores RecentVendorPurchases again, a filter the PR 75 merge had lost, and now MiscAccountData too. Three upstream fixes need rework in the fork's layout and follow separately: picking the most complete blob candidate (843c437), trade dialogs split across log reads (2048b64) and the low-confidence reward retry gate (b1f762e).
A stale copy of the inventory blob can stay intact in memory next to the live one. The walk sent whichever copy finished stitching first, so a stale copy at a lower address could roll the inventory back. The walk now collects every candidate and sends the one with the most items. After the first candidate it keeps reading a bounded stretch of memory to catch a second copy. Ports upstream 843c437.
The log watcher only kept a trade dialog when the whole dialog arrived in one read. When the game was still writing it as the tailer woke, the trade was never recorded. The watcher now keeps a rolling window of recent log text and parses the last dialog once the success line arrives. The game sometimes appends an icon glyph to the trading partner's name. It is now stripped, so the name matches the whisper sender. Ports upstream 2048b64 and the name fix from f709769.
A reward card could pass the score threshold on one or two OCR words, which cannot tell "Wisp Prime Neuroptics" from "Wisp Prime Chassis". Such a card is now low confidence unless its icon confirmed it, as is a card the catalog could not match at all. The overlay then waits for the soft retries before it locks the result. Upstream only replaces the best read on a higher card count, so its retries could never change the shown result. Here a confident read of as many cards replaces a low-confidence one. Ports the gate from upstream b1f762e.
* fix: complete WFM set sales as transactions * fix: refresh item catalogue on startup * fix: make WFM sale recording opt-in * fix: avoid duplicate catalogue startup refresh * fix: refresh catalogue daily without a second startup probe
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
…fix; bump to 4.4.13 Kitguns (Pistols/LotusModularSecondary*) shared a generic wrapper ItemType across all variants and were never given the same "key by defining component" treatment already used for Amps and Zaws. Every owned Kitgun collapsed onto at most 3 shared map keys with amount overwritten (not accumulated) and no resolvable name. Now keyed by the Barrel component path, same pattern as Amps/Zaws. Also includes the component-icon fix from wfcd.rs (parts/blueprints were showing their parent item's icon since WFCD dropped components[].imageName upstream; now resolved via ExportResources icon lookup by uniqueName). Version bumped to 4.4.13 across package.json, Cargo.toml, tauri.conf.json. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Upstream keeps the trade dialog parser and `handle_trade_completion` in log_watcher.rs. The fork now carries the same code there, and trade_log.rs keeps only the trade commands.
Blob capture runs from `spawn_blob_capture_thread` in its own module, and monitor.rs gains the catalog, startup-state and thread helpers under upstream's names. Behaviour is unchanged: the fork keeps its probe/walk scheduling and inventory cache model.
ocr/mod.rs keeps the shared reward matching, and ocr/linux.rs holds the X11 capture and Tesseract code behind upstream's function names. ocr/windows.rs is upstream's file unchanged and is not built. `OcrParams` moves to the crate root, where upstream keeps it.
The reward tailer in `start_monitor` and the always-on `start_log_watcher` both read EE.log and both emitted `wfm-whisper`, so each whisper showed up twice while the monitor ran. One thread, spawned at startup, tails EE.log for the app's lifetime from `reward_watcher.rs` and calls handlers in `log_watcher.rs` under upstream's names. A panic in one handler skips that chunk and the thread keeps running. The relic-reward OCR trigger fires only while the monitor is on; the riven unveil event fires with it off too. `start_monitor` moves to lib.rs.
Debug builds keep settings, database, caches and logs in .frameforge-dev/ next to the project instead of the installed app's directories, and are titled FrameForge Dev so the two environments are always distinguishable. The root resolution, window title and update-check skip compile out of release builds; production keeps using the same paths as before.
scan_log.txt was added to LEGACY_CACHE_FILES in v4.1.0, so the migration deletes it by design and the emptied legacy directory is removed. Seed an unrecognised file instead: that is the case the test exists to protect.
Amps, zaws and kitguns are keyed by their component path (prism, tip, barrel) when the inventory cache is built, but the change-detection state keyed them by the generic wrapper item_type. The cache therefore seeded a quantity the first scanned blob could not reproduce, so every app start reported the seeded count dropping to zero (Mote Prism 1 -> 0 was logged once per session) even though nothing was consumed. Extract inventory_state::modular_component_path() as the single source of keying for both, replacing the three duplicated blocks in build_inventory_from_blob.
fix: modular items reported as flipping to 0 on every scan
chore: isolate debug builds from production data
…ents WFCD hasn't picked up these new item paths yet, so add manual name corrections so they display correctly in the catalogue in the meantime. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
WFCD does not list these item paths yet, so the bundled corrections add them until it does.
Undo re-creates a fully sold listing, and warframe.market rejects a mod order that has no rank. The listing's rank now travels with the undo info.
With the new "Record automatic sales" setting on, a detected sale closes the order through warframe.market's close endpoint, so the sale counts in the seller's statistics. The setting is off by default, and FrameForge cannot undo a recorded sale. EE.log lists a traded set as its parts. The matcher now tries the listing named in the buyer's whisper first and falls back to the parts.
A development build reads and writes the installed app's settings, database, caches and warframe.market login. Setting FRAMEFORGE_ROOT moves all four roots under that directory. That run also stores its own login, keeps its factory-reset marker to itself, titles the window "FrameForge Dev" and skips the launch update check. Without the variable nothing changes.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Release v4.5.0
The merge keeps the fork's tree and takes only the version. Earlier fork commits already ported sale recording, the part name corrections, the mod rank on undo and the separate dev data root, the last one as opt-in. The fork already covers catalogue ETags, Kitgun keying, modular quantities and the debug file removal, and upstream's blueprint icon fallback matches no item in the current catalogue.
Hand-ported and skipped upstream changes left their commits outside main, so GitHub kept counting the fork as behind. A sync now always ends with the tag merge, and upstream's tags are fetched under refs/upstream-tags/ so their names no longer clash with ours. Fork-delta: fork-only
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.