Previously LEDGER_FILE=foo hledger add did, but hledger -f foo add didn't.
Now they both consistently will error if given a glob
(a path contining [, {, *, or ?) that matches nothing,
rather than auto-creating a file with a glob-like name.
Hledger.Utils.IO:
expandPathOrGlob
This restores the pre-1.50.3 behaviour of add and import, which once
again auto-create a missing file (specified by -f or LEDGER_FILE or
the builtin default path) rather than giving an error.
This fixes#2514 and refines the fix for [#2485].
There's also an improvement: they no longer create it unconditionally at the start;
they create lazily, when they have data to write.
Hledger.Read:
defaultExistingJournalPath
defaultExistingJournalPathSafely
readPossibleJournalFile
Hledger.Cli.Utils:
withPossibleJournal
Replace O(n log n) re-sorting of all prices on every valuation date
with O(log n) indexed lookups. By pre-building sorted price indexes
once at startup using O(n log n) time, we avoid redundant work
during reports.
This significantly improves performance for --value=end,COMM with daily
reports over long periods and large price databases.
Implementation:
- PriceIndex maps commodity pairs to a Map from date to effective
price, enabling O(log n) temporal lookups via M.lookupLE.
- DefaultValuationIndex provides efficient resolution of destination
commodities using the same temporal logic.
- makePriceGraph is updated to consume these indexes.
Signed-off-by: Oleg Bulatov <oleg@bulatov.me>
The nightly tag is no longer usable, having been used in a test of
immutable releases on Github. The branch/tag/workflow/release is
now "testbin", intended for publishing unreleased binaries for testing.
When including a literal path, don't use the Glob library at all.
Glob seems to read attributes of all files in a directory,
which disturbs build tools like tup which detect dependencies
based on filesystem operations.
Since 1.50.3, canonicalizePath was being called wastefully when
processing journals with many nested include files and/or many matches
for include glob paths. On a slow filesystem, with unusually
many includes, this might have been quite noticeable.
Now we canonicalise each file path just once as it is encountered,
avoiding the wasted IO work.
If transactions on the same date are coming from two files specified
with -f options, we expect them to be displayed in parse order, ie
respecting the order of the -f options. This wasn't always the case,
now it is.
Also, transactions' tindex field is now unique across all files,
where previously it started at 1 in each file. This affects hledger
data generally, not just the aregister command.
