Rework xaccSplitDetermineGainStatus() to resolve the gain status of
both ends of a marker/source pair in a single call, and to use the
xaccSplitGetGainsSourceSplit() accessor instead of a raw KVP lookup.
When the function resolves one split it now also resolves its partner
(the source for a marker, the marker for a source) if that partner is
still UNKNOWN. This makes gain-status resolution order-independent:
whichever split xaccSplitComputeCapGains() reaches first, the other is
already resolved, so the marker->source switch-over no longer finds an
unresolved source and never re-creates an orphaned gains split.
This supersedes the one-line switch-over patch in cap-gains.cpp, which
is reverted.
Builds a FIFO lot (buy + sell) with a pre-existing Realized Gain/Loss
transaction and the gains-split/gains-source KVP links, then simulates
a just-loaded state where the gains marker reaches
xaccSplitComputeCapGains() before its source is resolved (the order
that reproduces on MySQL and is masked on SQLite). Asserts the existing
gains split is reused rather than duplicated into an orphan.
Fails (2 != 1) without the switch-over resolve fix, passes with it.
xaccSplitComputeCapGains() performs a marker->source switch-over when a
gains marker split is processed before its source. The switch-over did
not resolve the source's in-memory gain status, so the source was left
with gains==UNKNOWN and gains_split==nullptr even though its forward
gains-split KVP was intact. The subsequent lookup read nullptr and the
code created a brand-new gains transaction routed to Orphaned Gains,
duplicating hundreds of already-settled gains splits on SQL books whose
history contains a stock split.
Resolve the source's gain status immediately after the switch-over so
its existing gains split is reused instead of being re-created.
This latent bug (original cap-gains code, ~2003) was exposed by the
5.17 adjusted_amount work (1e654412ab, b1efe0763c) which recomputes
already-booked gains in stock-split accounts. It is order-sensitive,
which is why it reproduces on MySQL but not SQLite.
A templated account->splits iterator, will target splits from
start_date to end_date, inclusive of date boundaries. Also optionally
include descendants.
gnc_price_set_source_string maps a source string to the PriceSource enum and
silently leaves the source unchanged for anything it doesn't recognize -- a
silent no-op a naive Python or Guile script can hit. Steer callers to
set_source() with a PriceSource member (added in #2317) instead.
Python bindings:
- Deprecate GncPrice.set_source_string(): it now emits a DeprecationWarning
pointing at set_source(PriceSource.<...>). It still works during the
deprecation period and still guards -- if the string is not recognized (the
round-tripped source no longer equals the input) it raises ValueError rather
than silently doing nothing.
- Fix the deprecated() decorator in deprecation.py: it referenced the undefined
DeprecationWarnig and inspect.current_frame(), so it raised instead of
warning (the one existing @deprecated use was latently broken).
Guile bindings:
- mark gnc-price-set-source-string deprecated in engine.i.
(Python and Guile use separate SWIG interface files, so Python is
handled separately). The only in-tree Guile caller is the QIF importer, updated to
gnc-price-set-source with PRICE-SOURCE-USER-PRICE; the redundant
set-source-string in test-engine-extras.scm's fixture helper is dropped.
Tests (test_price_and_wrapping.py):
- TestSetSourceStringGuard covers the guard (still raises on an unrecognized
string) and the new DeprecationWarning.
Removing set_source_string from the Python bindings is a follow-up.
PR #2187 made GncPrice.get_value() return a GncNumeric wrapper and clone() a
fully wrapped GncPrice. The price example scripts still used the pre-5.15 raw
_gnc_numeric idiom and broke -- reported as Bug 799796 (cloned prices silently
kept the original value).
Example scripts (bindings/python/example_scripts/):
* quotes_historic.py -- reworked to the modern idiom; it now actually runs on
5.x (it had further latent breakage beyond the reported one):
- Construct each price directly with GncPrice(book) + setters instead of
cloning an existing price, so the script no longer needs a pre-existing
price in the book. Look the pricing currency (USD) up from the commodity
table rather than deriving it from a split.
- Build the value with the double->GncNumeric constructor at GnuCash's
standard price precision -- the pricing currency's SCU * COMMODITY_DENOM_MULT
rounded HALF_UP (gnc-pricedb.h "Price policy") -- dropping the
Fraction.from_float().limit_denominator(100000) dance and its import.
- Set the source with the PriceSource enum -- set_source(PriceSource.FINANCE_QUOTE)
(the enum added in #2317). The original set_source("Finance::Quotes::Historic")
passed a string to what has been an enum setter since 2.6, which raises
TypeError on modern bindings.
- Open with Session(url), which defaults to SESSION_NORMAL_OPEN. The old
Session(url, True, False, False) passed the deprecated ignore_lock=True,
silently bypassing the book's lock file.
* priceDB_test.py, price_database_example.py -- num/denom are methods now, so
read v.num() / v.denom().
Prevent silent recurrence (gnucash_core.py):
The write path failed *silently*, which is how this slipped through review and
CI. num/denom are read-only accessor methods, so assigning to them merely
shadows the method and changes nothing. Add a GncNumeric.__setattr__ that raises
AttributeError on assignment to num/denom, pointing at the correct idiom
GncNumeric(num, denom).
Test (bindings/python/tests/test_numeric.py):
New test_num_denom_read_only covering the guard. It needs no backend, so it runs
in the existing python-bindings CTest on all platforms.
The wiki page Stocks/get_prices, which the reporter copied from, is being
updated separately.
Setting a price's source from Python meant reaching into gnucash_core_c for the
raw PRICE_SOURCE_* constants. Add a PriceSource(IntEnum) alongside SessionOpenMode,
with values sourced from the C enum (via gnucash_core_c) so they can't drift, so a
script can write price.set_source(PriceSource.FINANCE_QUOTE) with a plain top-level
import.
Add test_price_and_wrapping.TestPriceSourceEnum:
- test_values_match_c_enum asserts the Python enum covers the C PRICE_SOURCE_*
value set exactly, so adding, removing, or renumbering a C value fails CI until
PriceSource is updated to match. (Member names are a Python-side choice --
FINANCE_QUOTE spells out FQ -- so the check is on values.)
- test_set_source_accepts_enum confirms a member flows through set_source() and
round-trips.
Reorder the CLI quotes get startup sequence to load GncQuotes before
loading the session so that preconfigured quote sources don't null
out other supported sources.
Make the SQL backend create a new source when an unknown one is found in
the commodities table like the XML backend does.
Thanks to Ganesh Suriyanarayanan for diagnosing the problem and
proposing a solution.
John Ralls
Pedro Albuquerque
Ceaus
Brent McBride
Yegor Derevenets
zugu
Alois Levermann
Szia Tomi
Arve Eriksson
Jonathan Santos Fernandes
Simon Arlott
Zhaoquan Huang
Christopher Lam
Noerr