Repository navigation
Engine changes for 1.14.0 - #912
Conversation
The timezone bullet in the 1.14.0 notes began its second sentence with "Paired codegen 1.14.0 also makes calendar-form" and went on to say that timestamp() stops at its first evaluation when the engine rejects a zone name. The paired codegen release is 1.13.0, the release that stopped the silent UTC fallback of codegen 1.12.0. The bullet now says 1.13.0. Fourteen documentation lines that the combined change re-added named private fixture sets, private tapes and private tests by file or test name. Each now says "a private fixture set" or "a private test" and keeps the rest of its sentence: - docs/design/native-feature-parity.md, rows SZ3, MG3, MG4, MG7, MG15, FP4, RP1 and OT7; - docs/adr/0001-kernel-adapter-boundary.md, the rows `present_session_day` and the one for a daily bar the installed daily feed dates; - docs/pages/pine-to-native.md, the rows `pyramiding`, `session.ismarket`, `session.isfirstbar` and `session.islastbar`. The parenthesis in the `session.ismarket` row that cites tests/test_session_ismarket_tape.cpp also carried the internal work label of the change that wrote the test. The label is dropped and the citation stays. Citations, counts and every other word of those lines are unchanged.
The stamp that pineforge-live checks before it reads a module's engine handles
stayed at 4 while the native consumer's layout changed: the live runner's
memory change inserts three counters before members the runner reads and
writes. The stamp is now 5, so a strategy module built against 1.13.x headers
declares 4 and is refused at admission ("strategy engine layout mismatch:
module=4 runtime=5") instead of being read at shifted offsets.
- include/pineforge/pineforge.h, src/c_abi.cpp and
tests/test_abi_layout_probe.cpp pin the value 5.
- tests/native_live_startup_e2e.py expects stamp 5 from the current example
module and runtime=5 in the refusal of the stamp-2 control, and runs a
second control: tests/test_live_contract_prior_release_layout_plugin.cpp
declares 4, the stamp a 1.13.x module carries. runner/CMakeLists.txt builds
it as a module and registers native_live_prior_release_layout_e2e, which
asserts the refusal line, no output and no ledger, and prints the line.
- runner/README.md says that modules declaring a lower stamp must be rebuilt.
- docs/design/native-feature-parity.md cites test_native_example_batch at its
new line in runner/CMakeLists.txt (the new module target moved it by six).
- CHANGELOG.md: one compatibility entry. Modules built against 1.13.x headers
are refused and must be rebuilt. The earlier miss is disclosed: 1.13.0
changed the same internal layout without raising the stamp, so a module
built against 1.12.x headers and loaded into a 1.13.x runner was not refused.
PF_ABI_VERSION stays 5; the C API and its caller-allocated structures are
unchanged.
…ted chart On a chart aggregated from a lower timeframe, a confirmed bucket's decision floor is raised to the completion of its inputs before the bucket's modeled path is walked, while the path's open point is dated at the bucket's first input. An order that a fill places or replaces at that open point (the exits of an entry that fills there, or the exits a pyramiding add re-arms) was therefore treated as born at the floor, the bucket's close: the segments of the same bucket could not fill it, and an exit that the bucket crosses after its open filled on a later bar or point, with a different drawdown and run-up than the plain chart gives. The open point of the bucket being walked now counts as current for a request made in its fill callback, as the plain chart's open point does: a file-local helper and a private member, applied_open_is_current, that make_command_context calls in place of its first floor test. Both are defined at the end of the file, so that no line cited by the documentation moves. The test covers every request made in such a callback, so the exits of a first entry change as well as an add's. Plain charts, every other path phase and every other provenance are unchanged. The change reaches an aggregated chart only. applied_open_is_current also requires aggregates_input() (src/native_execution_consumer.hpp:262, the accessor behind native_aggregates_input_bars), which is native_calendar::pairing_aggregates of the pairing the run configured from its input and script timeframes (pairing_, src/native_execution_consumer.hpp:1953, set at src/native_execution_consumer.cpp:2889 and :3097; src/native_calendar.cpp:1664). That predicate is false for Passthrough, the byte-identical timeframes of a plain chart (include/pineforge/native_calendar.hpp:100), and for a run whose timeframe was not detected, whose pairing keeps its default. On a plain chart the helper is therefore the first floor test alone, as before this change, and a request keeps the birth it had. The condition is the last test of applied_open_is_current (src/native_execution_consumer.cpp:15339). tests/test_aggregated_add_exit_leg_birth.cpp pins two constructions (ten-minute bars aggregated from minute bars against the same bars given directly): the exit an add's fill re-arms, and the take-profit a first entry's fill places that its own bucket crosses. tests/test_aggregated_chart_add_exit_differential.cpp runs seeded windows of minute bars through both constructions for seven shapes (no add, an add with its own lot, two ids, a same-id add, a limit-priced add, a stop-priced add, and a first entry whose take-profit its own bar crosses) and asserts identical closed rows.
….kcw The fixed-length classes ta::EMA, ta::TSI, ta::MACD, ta::KC and ta::KCW take any length. A zero or negative length gives alpha = 2.0 / (length + 1) values or a division by zero, where TradingView stops the run with an error. pineforge/source/pine_ta_length.hpp now has PineEma, PineTsi, PineMacd, PineKc and PineKcw. Each takes the arguments of its class, stores the lengths and never stops at construction, so an invalid length cannot stop a run before its first bar. The first compute() or recompute() checks every length with simple_ta_length, in the signature's order and with TradingView's function and argument names: an na length, zero or a negative length stops the call with pine_invalid_argument and TradingView's text. A passed check builds the fixed-length class, and every call then delegates to it, so for a valid length every value is the class's own. A failed check builds nothing and the next call checks again. The ambient ta::ema_na_warmup_flag() is neither read nor changed. No existing behaviour changes: the fixed-length classes are untouched, and code generated by the released code generator does not reference the new types. The code generator's paired change routes its EMA-family calls through them. The new test compares each type with its class bit for bit over 48 bars with na gaps (compute, and recompute of the same bar), checks the refusal of na, 0 and negative values for every length argument (no stop at construction, the stop at the first compute and at the first recompute, the exact text and code, the second call failing again, the bar prefix), and records that the fixed-length EMA takes length 0 and computes. For every type it also runs the shapes that generated code uses: a recompute as the very first call, the reset `member = Type(args)` after use (to a valid length, and to an invalid one and back), and copies taken between calls. Refs pineforge-4pass/pineforge-codegen-oss#291.
A request for a fixed period that is not a multiple of the chart, such as "75" on a 60-minute chart, was served from the chart's own bars whenever the integer ratio of the two timeframes rounded down to passthrough. A script that sized its entries from the requested close then booked the wrong quantities. When bars of the requested period are supplied, the evaluator now keeps a fixed-period clock. Each requested bar completes at its own close, clipped by the session, and is published on the chart bar that close belongs to: on the chart-input, finer-input and auxiliary-feed routes, with lookahead and with gaps. For an exchange instrument, the last input of a session-day that closes early ends the period it leaves open. A supplied bar that is off the requested grid is refused. A native subscription of such a period is admitted when its bars are supplied. Requests the chart divides, calendar periods and requests without supplied bars keep their behaviour. The clock and the per-chart gap marker are part of the state hash. With lookahead, the evaluator reads the supplied bars through the feed index the run's routing chose. The host's feed setter stays usable during a run, and a change to the feed list after the routing can leave that index on another timeframe's feed or past the list. Every change to the list now marks it, the routing clears the mark, and while it is set the read refuses the run with "native security feeds changed during the run" instead of reading a feed the routing did not choose. A run started on a handle that is not ready is now refused with "native run requires configure_native before each run". The run-failure catalog lists the new text beside the earlier one, and the documentation says that every run needs its own configure. The source allocation-bound manifest lists the evaluator's input timeframe text as one new open group, and five new site identities in existing open groups; each entry says what limits its site. The documentation's line citations follow the source lines this change moves. Refs #718
A native subscription of a fixed period that the input does not divide, such as 75 minutes over a 30-minute input, is admitted when its bars are supplied. When one of those bars does not open on the requested grid, the run is refused at begin with "native fixed-period bar is off the requested grid". The refusal names the code security_feed_refused with the reason native_feed_grid and the requested timeframe, but the run-failure catalog listed neither that reason nor that argument, so the run read engine_invariant with no arguments. The catalog now lists native_feed_grid among the reasons of security_feed_refused, an optional requested_tf argument of the timeframe kind, and the refusal's text. The registry, the catalog diff and the code table of the documentation are regenerated from it. No code is added or renumbered, and the generated code list does not change. Two citations of later catalog lines in the binary input design page follow the lines this change moves. A native subscription test runs a 30-minute input with supplied 75-minute and 105-minute bars, the second of them one minute after its grid open, and reads the refusal through the public getters: its text, the code security_feed_refused, and the reason native_feed_grid with the timeframe of that request. The same bar on its grid open is delivered as supplied and the run completes. Refs #718
The fixed-period request change moves the citation numbers in several rows of
the kernel/adapter boundary table and the native feature table, which makes the
whole row new text. Four of those rows named the reference captures of
TradingView's output with the name of an internal command, and a fifth carried
two labels of the design work's stages.
The four rows now say "the TradingView reference capture" (or "captures") where
they named the command, and the fifth drops its stage labels and keeps the hook
they named ("sub-bar hook"). No citation number, file name, test name or other
word changes.
One of the rewritten sentences is the one that b97c409 ("Remove the
non-distributed test inputs from the public tree (PR 1 + PR 1b) (#371)")
introduced in the native feature table, in the row for the auxiliary feed
driving request.security: its first words, which named the reference captures
by the name of an internal command, are replaced; the rest of the sentence,
"the bar's first recalculation before its close callback feeds it", is
unchanged. This message names it so that the documentation revert check can
tell a rewording from a deletion.
There was a problem hiding this comment.
Graphify code review 
Review completed · 4539ad1b55cd
Warning
Merge confidence: 3/5 · Review before merging · 🟠 High impact
1 unranked bug checked against the code · 19 possible issues not confirmed
The change adds fixed-period completion for native higher-timeframe series and introduces checked adapters for several Pine TA functions.
Full summary
The change adds fixed-period completion for native higher-timeframe series and introduces checked adapters for several Pine TA functions. It also changes native run lifecycle requirements and adds per-call-site currency-conversion warnings.
Bugs (1)
Cap simple_ta_length at INT_MAX - 1: ta::EMA computes length + 1, which overflows when PineEma::bound() passes INT_MAX. Inspect source
Detailed review · checked against the code
Very large EMA lengths overflow while building the adapter
A length above INT_MAX is clamped to INT_MAX, whose increment overflows when the EMA is constructed.
A direct call to PineEma::compute or recompute reaches bound(), which passes the stored length through simple_ta_length before constructing ta::EMA. That helper clamps a positive double or int64_t length above INT_MAX to INT_MAX. EMA then evaluates length + 1 as signed int in its constructor; at INT_MAX that addition overflows, which is undefined behavior in C++. Thus the claimed adapter path can construct an EMA with an invalid coefficient calculation for an oversized length.
- Independent source review: The adapter test exercises ordinary valid lengths and rejects non-positive lengths, but the inspected test section does not cover oversized positive lengths. The changelog says released codegen does not reference these new adapter types, so this does not establish that released generated strategies reach the issue; direct callers of the adapter do reach it through its compute/recompute methods.
Not confirmed (19 possible issues)
Leads worth a look; we didn't double-check these. They don't change the score.
Not confirmed · Fixed-period replay reads kBars at index -1 when a chart bar has no requested value, which is undefined behavior Inspect source
Detailed review · not confirmed
The trade check indexes the fixture bar table with the expected row for each scheduled chart bar, without first confirming that the row exists.
Line 597 walks chart bars 7 onward at the script's entry cadence, and line 603 uses kBars[expected[i]] unconditionally. The probe books an entry only when latest.present is true (line 511), and expected[i] is -1 exactly when the read is absent (line 561). Nothing skips the absent case before the index.
-
Only the 60-minute, non-lookahead branch of this loop was read; the rest of the file was truncated.
-
Found by the per-file sweep; not double-checked.
Not confirmed · std::max and std::min are used without including <algorithm>, so the file depends on transitive includes Inspect source
Detailed review · not confirmed
The aggregate helper calls std::max and std::min, but the file does not include <algorithm>.
The includes shown at lines 28-35 cover cassert, cmath, cstdio, cstdint, limits, string and vector, but not algorithm. std::max is used at line 1108 and std::min at lines 1109 and 1136.
-
The full include chain of the included headers was not inspected.
-
Found by the per-file sweep; not double-checked.
Not confirmed · In NDEBUG builds the test's assert checks are compiled out, so value mismatches and missing refusals pass silently and ok prints anyway Inspect source
Detailed review · not confirmed
Every check in this file uses assert, including the bit comparison in same and the requirement in stops that a call throws. With NDEBUG defined these checks vanish, so the test can report success without verifying anything.
caught is only checked by assert(caught) on line 122. same checks bits(actual) == bits(expected) only through assert on lines 45, 48 and 49. main unconditionally prints "ok" on line 464. The file contains no non-assert check, so nothing else would catch a failure.
-
The build flags are not shown, so this applies only if NDEBUG is defined for this target
-
No CMake or build configuration was inspected
-
Found by the per-file sweep; not double-checked.
Not confirmed · A non-RunFailureInfo exception leads to a null dereference on line 118 in NDEBUG builds instead of a clean test failure Inspect source
Detailed review · not confirmed
stops casts the caught exception to RunFailureInfo and checks the pointer only with assert. If the cast gives null and asserts are disabled, the next line dereferences it.
Line 117 is assert(failure != nullptr), and line 118 dereferences failure immediately. The guard disappears in release builds, so nothing stops the null pointer from being used.
-
Only applies when the thrown type is not RunFailureInfo and NDEBUG is defined
-
Found by the per-file sweep; not double-checked.
Not confirmed · Trade rows may pair the wrong closed-trade records if open trades are counted in the loop Inspect source
Detailed review · not confirmed
rows() loops over trade_count() but reads each row through the closed-trade accessors using the same index, so the two index spaces may not match.
Lines 190-196 call get_trade(i) and the closed_trade_*(i) accessors with the same i, and the Row type has an open_at_end field, so open trades are expected in this output. The shown code does not define whether trade_count() counts closed trades only, so the indexing is unverified.
-
The definitions of trade_count, get_trade and closed_trade_* are not in the shown code, so whether they share one index space is unchecked.
-
Found by the per-file sweep; not double-checked.
Not confirmed · Five identical limit/requested/used argument rows are added for source location 369, so failure details may repeat or be ambiguous Inspect source
Detailed review · not confirmed
Rows at HEAD lines 686-700 repeat the same limit, requested and used string entries five times for the same location.
The arg table is indexed by location and the rows at 686-700 are identical copy-paste duplicates. Only one set of these fields should belong to that failure, so the extra rows either leak into the output or cannot be told apart.
-
The reader of this table is not in the shown code, so I could not confirm whether it dedupes rows or reads them by count.
-
Found by the per-file sweep; not double-checked.
Not confirmed · Location 533 carries about 23 unrelated argument names, such as fill_ms, pruner and position_size, which may be attributed to the wrong fail Inspect source
Detailed review · not confirmed
Rows at HEAD lines 730-753 all point to location 533 and mix fields from different failure kinds, with some names repeated.
Many unrelated fields (for example fill_ms, pruner, resting_orders and timeframe) share one location with count 0 in the adjacent rows, so the table cannot be matching each field to the code that uses it.
-
The indexing scheme is not fully visible, so I could not confirm that these rows are misattributed rather than intentionally grouped.
-
Found by the per-file sweep; not double-checked.
Not confirmed · Pre-session bars in charts are bucketed into the first session chart bar instead of their own earlier bucket Inspect source
Detailed review · not confirmed
charts truncates toward zero for minutes before the session open, so pre-open bars land on the 09:30 bucket.
At line 170, (minute - kOpenMinute) / chart * chart uses C++ integer division, which truncates toward zero. For a bar at 09:29 with chart=5, (-1)/5*5 is 0, so the open is 09:30 instead of 09:25. The helpers in this header only generate session bars, so this path is not exercised here, but the function is still wrong for pre-session input.
-
Only inputs with minute < 570 (before 09:30) are affected; the
inputandbarshelpers never produce them. -
Found by the per-file sweep; not double-checked.
Not confirmed · day_of gives the wrong session-day for timestamps before the origin Inspect source
Detailed review · not confirmed
day_of divides a negative offset with truncation toward zero, so times just before the origin map to day 0 instead of day -1.
Line 57 computes (ms - kOriginMs) / (kDayMinutes * kMinuteMs). For a negative numerator, C++ integer division rounds toward zero, so -1 ms gives 0 rather than -1. Every caller that groups by day_of inherits this.
-
Only timestamps earlier than kOriginMs are affected; the fixture never builds bars before the origin.
-
Found by the per-file sweep; not double-checked.
Not confirmed · EMA-family adapters build their fixed-length object on the first bar, so the global EMA warm-up flag is read then, not when the adapter is c Inspect source
Detailed review · not confirmed
The adapters store the lengths in the constructor and build the ta::EMA, ta::TSI, ta::MACD, ta::KC or ta::KCW object lazily, inside the first compute() or recompute(). The header says the built class reads the ambient ta::ema_na_warmup_flag() as it always does, so if that read happens at construction, the value seen is the one current at the first bar.
Lines 129-130 say the flag is neither read nor changed by these adapters and is read by the built class as it always does. Lines 147-150 show bound() emplacing ta::EMA only when compute() or recompute() first runs. Before this change the fixed-length classes were presumably constructed where the code declared them, so the flag was read earlier. The header does not show where the flag is read, so this timing change is not confirmed.
-
The header does not show whether ta::EMA and the other fixed-length classes read ema_na_warmup_flag() at construction or on each compute call.
-
Whether any code changes the flag between adapter creation and the first bar is not visible in the given code.
-
Found by the per-file sweep; not double-checked.
9 more possible issues are available in Graphify.
What this review covered
- We didn't review 23 files that aren't source code, such as docs, config, images, or generated or third-party code.
- One review pass used all of its turns for reading code, so it reported from what it had already read.
- A changed file's diff was too long to show in full, so the review may not have seen every changed line in it.
- 38 of 38 changed files inspected. This is scope, not a correctness guarantee.
Code graph and static checks · 10963 functions depend on this change
Impact & health
Impact — 10963 functions depend on the 6020 functions this change touches.
Health — this change adds coupling hotspots:
- worse:
tf_change()— 28 callers, 5 callees - worse:
crosses_boundary()— 8 callers, 8 callees - worse:
native_run_spec_digest()— 17 callers, 3 callees - worse:
build_native_day_partition()— 9 callers, 5 callees - worse:
outcome_of()— 11 callers, 4 callees - worse:
session_covered_instant_ms()— 14 callers, 3 callees - worse:
intraday_bucket()— 7 callers, 5 callees - worse:
make_command_context— 2 callers, 12 callees - …and 15 more — each is listed as a finding
Formal checks — 10963 functions this change can affect weren't formally checked in this review. Formal checks are informational only.
Gate & verification
✅ PASS — objectively clean (no health regressions, tests not run — proofs not run this pass (advisory)). Grounded, not self-assessed.
Advisory (not blocking):
- verification_scope: 10558 function(s) in the blast radius were not formally verified this run
Test selection
1465 of 1465 test file(s) selected (100%) via static blast radius.
Escalated to a full run for safety (
noncode-change,uncovered-code) — the selection is not trustworthy on its own (see below). CI should run the whole suite.
benchmarks/tests/test_aggregate.py— full-run-safetybenchmarks/tests/test_check_provenance.py— full-run-safetybenchmarks/tests/test_compare.py— full-run-safetybenchmarks/tests/test_public_names.py— full-run-safetybenchmarks/tests/test_run_all.py— full-run-safetybenchmarks/tests/test_run_strategy_feeds.py— full-run-safetydocker/pineforge_sink_finalize_test.py— full-run-safetydocker/pineforge_sink_replay_rq_test.py— full-run-safetydocker/pineforge_sink_replay_test.py— full-run-safetydocker/pineforge_sink_to_json_test.py— full-run-safetydocker/quota_columns_test.py— full-run-safetydocker/quota_convert_test.py— full-run-safetydocker/quota_finalize_test.py— full-run-safetydocker/replay_rq_engine_test.py— full-run-safetydocker/run_json_artifact_window_test.py— full-run-safetydocker/run_json_bin_test.py— full-run-safetydocker/run_json_bounded_transport_test.py— full-run-safetydocker/run_json_bulk_equity_fuzz_test.py— full-run-safetydocker/run_json_bulk_equity_test.py— full-run-safetydocker/run_json_codes_test.py— full-run-safetydocker/run_json_converter_metadata_test.py— full-run-safetydocker/run_json_converter_session_test.py— full-run-safetydocker/run_json_diagnostics_test.py— full-run-safetydocker/run_json_equity_scan_test.py— full-run-safetydocker/run_json_fused_test.py— full-run-safetydocker/run_json_harnesses_test.py— full-run-safetydocker/run_json_metrics_only_test.py— full-run-safetydocker/run_json_misc2_test.py— full-run-safetydocker/run_json_misc_test.py— full-run-safetydocker/run_json_pfbar_test.py— full-run-safetydocker/run_json_provenance_test.py— full-run-safetydocker/run_json_quota_supervisor_test.py— full-run-safetydocker/run_json_replay_test.py— full-run-safetydocker/run_json_selected_sink_test.py— full-run-safetydocker/run_json_sink_lexical_test.py— full-run-safetydocker/run_json_sink_summary_test.py— full-run-safetydocker/run_json_sink_test.py— full-run-safetydocker/run_json_sink_vectors_test.py— full-run-safetydocker/run_json_spool_work_test.py— full-run-safetydocker/run_json_stringio_test.py— full-run-safetydocker/run_json_symbol_feeds_test.py— full-run-safetydocker/run_json_syminfo_test.py— full-run-safetydocker/run_json_trade_scan_test.py— full-run-safetydocker/run_json_window_budget_test.py— full-run-safetydocker/run_json_wrapper_work_test.py— full-run-safetyinclude/pineforge/native_run_spec.hpp— full-run-safetyscripts/derive_corpus_feeds_self_test.py— full-run-safetyscripts/fingerprint_self_test.py— full-run-safetyscripts/run_corpus_self_test.py— full-run-safetyscripts/test_abi_receipt_skips.py— full-run-safety- … and 1415 more
⚠️ non-code file(s) changed (CHANGELOG.md,ci/cr150-source-bounds.yaml,ci/cr150-source-sites.json,docker/run_failure_codes.json,docker/run_failure_codes_diff.json…) → running the full suite for safety (a code graph can't see config/fixture/data deps)
⚠️ changed code file(s) with no mapped test (CHANGELOG.md,docs/adr/0001-kernel-adapter-boundary.md,docs/coverage.md,docs/design/BARS-BIN-INPUT.md,docs/design/arena-bound.md…) — a coverage gap or a missing link — running the full suite rather than only the selected tests
Selection is safe under the controlled-regression assumption; always-run tests + a periodic full run are the backstops. Advisory — it never changes the check verdict.
Docs that may be stale (advisory)
docs/design/o1-selected-capture.md§ Presentation hooks and compatibility (lines 109-147): references changed symbolsHostdocs/pages/abi-stability.md§ Symbol inventory (lines 149-281): references changed symbolsHostdocs/pages/abi-stability.md§ What's *not* guaranteed (lines 401-707): references changed symbolsHost
Every flagged function
tf_change()insrc/timeframe.cpp:1145· 28 callers, 5 callees: 28 callers depend on it (afferent coupling)native_run_spec_digest()insrc/native_execution_consumer.cpp:14961· 17 callers, 3 callees: 17 callers depend on it (afferent coupling)session_covered_instant_ms()insrc/timeframe.cpp:940· 14 callers, 3 callees: 14 callers depend on it (afferent coupling)outcome_of()intests/test_host_currency_conversion.cpp:121· 11 callers, 4 callees: 11 callers depend on it (afferent coupling)build_native_day_partition()insrc/timeframe.cpp:861· 9 callers, 5 callees: 9 callers depend on it (afferent coupling)crosses_boundary()insrc/timeframe.cpp:1083· 8 callers, 8 callees: fans out to 8 callees (efferent coupling); 8 callers depend on it (afferent coupling)make_command_contextinsrc/native_execution_consumer.hpp:1618· 2 callers, 12 callees: fans out to 12 callees (efferent coupling)timeframe_time_close()insrc/session_time.cpp:1896· 2 callers, 12 callees: fans out to 12 callees (efferent coupling)TimeframeAggregator::next_fixed_completion()insrc/timeframe.cpp:1249· 0 callers, 13 callees: fans out to 13 callees (efferent coupling)main()intests/test_checked_ema_family.cpp:449· 0 callers, 12 callees: fans out to 12 callees (efferent coupling)intraday_bucket()insrc/timeframe.cpp:352· 7 callers, 5 callees: 7 callers depend on it (afferent coupling)refusal()intests/test_checked_ema_family.cpp:144· 4 callers, 3 callees: high coupling complexity (Ca·Ce = 12)session_in_postmarket()insrc/session_time.cpp:1679· 3 callers, 4 callees: high coupling complexity (Ca·Ce = 12)session_in_premarket()insrc/session_time.cpp:1644· 3 callers, 4 callees: high coupling complexity (Ca·Ce = 12)timeframe_time()insrc/session_time.cpp:1848· 2 callers, 9 callees: fans out to 9 callees (efferent coupling)session_window_close_ms()insrc/timeframe.cpp:1041· 2 callers, 8 callees: fans out to 8 callees (efferent coupling)generated_usage()intests/test_checked_ema_family.cpp:417· 1 callers, 10 callees: fans out to 10 callees (efferent coupling)macd()intests/test_checked_ema_family.cpp:311· 1 callers, 6 callees: fans out to 6 callees (efferent coupling)tsi()intests/test_checked_ema_family.cpp:295· 1 callers, 6 callees: fans out to 6 callees (efferent coupling)convert_identity_or_warn_sites()intests/test_host_currency_conversion.cpp:644· 0 callers, 8 callees: fans out to 8 callees (efferent coupling)main()intests/test_aggregated_chart_add_exit_differential.cpp:284· 0 callers, 7 callees: fans out to 7 callees (efferent coupling)BacktestEngine::hash_fixed_security_clocks()insrc/engine_state_hash.cpp:24· 0 callers, 6 callees: fans out to 6 callees (efferent coupling)reset_after_use()intests/test_checked_ema_family.cpp:202· 0 callers, 6 callees: fans out to 6 callees (efferent coupling)
Summary
Engine changes for the 1.14.0 release, as one squash.
ta.ema,ta.tsi,ta.macd,ta.kcandta.kcwstop a run with TradingView's error for an invalid length.ta::KCandta::KCWvalidate their constructor arguments. Code generated by the released code generator does not reference the new types.request.securitybars (Known issue in PineForge Engine 1.9.1: non-multiple requested timeframes #718). Completes fixed-timeframe requests on the native path, refuses a change of the feed list during a run, and classifies a supplied bar that is off its requested grid. The run-failure catalog gains one reason with an optional timeframe argument; no existing argument is renamed.Refs #376
Refs #291
Refs #718
Verification
Every change was proven failing on the base and passing with the change, and reviewed independently. The release, debug, sanitizer, kernel, documentation and parity checks run on this exact tree before merge, together with a paired parity sweep against the code generator changes for the same release.