Adds a Boost test that pre-creates a rocksdb/ dir with a
MIGRATION_INCOMPLETE marker and verifies the init path:
1. Detects the marker
2. Refuses to open the rocksdb/ dir as live state
3. Re-runs the migration
Also exercises the M4 marker flush+verify path: marker is on
disk only during an in-progress migration and removed on success.
This test was in the H1/H2/H3/M4 patch but never committed;
folding it in here so the test surface matches the audit doc.
H1: init.cpp now detects crashed migrations (MIGRATION_INCOMPLETE marker)
and retries instead of opening a partial RocksDB. Refuses to start if
the marker persists after migration attempt.
H2: LevelDB ExistsRaw now returns false for keys deleted in the active
batch, matching ReadRaw and the RocksDB backend. Fixes latent
cross-backend consensus split in intra-batch spend checks.
H3: All RocksDB close paths now go through close_rocksdb() which
destroys CF handles before deleting the DB. Fixes RocksDB assertion
/ UB on shutdown and version-reset.
M4: Migration marker write is now flushed + verified (refuses to start
migration if marker can't be written).
CF routing permanently disabled: GetCF() always returns nullptr (default
column family). The read path (NewIterator, LoadBlockIndex) only iterates
the default CF, so writes routed to per-prefix CFs were invisible to scans.
This is why -chaindb=rocksdb compiled clean but was never runtime-valid.
Existing CF-enabled DBs still open (handles retained for cleanup) but no
routing occurs. CF-aware iteration is a future follow-up.
txdb-factory: RocksDB is now the default backend (was still leveldb).
Tests updated for RocksDB-as-default expectations.
From Claude's ROCKSDB-T010-REVIEW-2026-07-01 audit on E:\repos\triangles.
hdPassphrase was hardcoded to empty string in DeriveHDKey, meaning users
who set a BIP39 passphrase during seed creation would derive different
addresses after restoration. This adds proper passphrase storage,
encryption, and decryption alongside the existing mnemonic handling:
- wallet.h: hdPassphrase + vchCryptedHDPassphrase + hdPassphraseIV fields
- wallet.cpp: Lock/Unlock/EncryptWallet/SetHDSeed all handle passphrase
with the same encrypt/decrypt lifecycle as the mnemonic
- DeriveHDKey now passes hdPassphrase to DeriveTriangles (not hardcoded )
- walletdb.h: WriteHDPassphrase/WriteHDCryptedPassphrase/EraseHDPassphrase
- walletdb.cpp: ReadKeyValue handles hdpassphrase/hdcpassphrase records
- rpcwallet.cpp: hdnew/hdshow show passphrase_used + warnings
Also: i2p.cpp hardens I2P private key file permissions to owner-only.
From Claude's uncommitted work on E:\repos\triangles (SAMI-PC). The rest
of Claude's modernization (Boost removal, RPC rewrite, RocksDB default,
SQLite wallet) was already committed to master in bfdb399 and follow-ups.
A v3 snapshot carries the UTXO set and raw block data (blk0001.dat) but
does NOT rebuild the per-tx index (txindex) that maps CTransaction hashes
to CDiskTxPos. Without it, any new PoS block referencing a pre-snapshot
transaction fails CheckProofOfStake with 'read txPrev failed':
CTransaction::ReadFromDisk(txdb, prevout, txindex) // src/main.cpp:714
if (!txdb.ReadTxIndex(prevout.hash, txindexRet)) // empty!
return false;
This stalls the node at the snapshot height and triggers DoS=100 on
every inbound peer feeding canonical blocks, masking as a network
misbehavior issue. DNS3 was stuck at 2,214,547 for this reason despite
the UTXO and block data being present.
Fix: after extracting blk0001.dat, walk it linearly and call
txdb.UpdateTxIndex(hash, CTxIndex(CDiskTxPos, nVout)) for every
transaction. O(N) over the historical block range, batched every
5000 txs. Adds ~30-60s to snapshot load on modern hardware.
This is the third leg of the v3 self-contained snapshot story:
v3 field source purpose
----------- ------------------------- -------------------------
headers last 2000 block headers block index continuity
utxos 22k unspent outputs UTXO set at tip
blocks blk0001.dat raw bytes on-disk block storage
setStakeSeen last 5000 PoS seen stakes stake collision dedup
txindex [this commit] PoS signature verification
Future 'v4 snapshot' work should consolidate all five into a single
load pass with progress reporting.
The 2026-06-30 commit cbb189a changed bnNewBlock > bnRequired to bnNewBlock < bnRequired,
but bnNewBlock is the candidate's compact-bits TARGET (not difficulty). In Bitcoin/PoS,
larger target = easier difficulty. The correct reject condition is when the block's
target is LARGER than required (i.e. block is easier than allowed for elapsed time):
bnNewBlock > bnRequired.
The inverted condition caused DNS3 to reject every canonical post-snapshot block as
'too little proof-of-stake' because most honest blocks satisfy bnNewBlock < bnRequired
(block is harder than the very-loose anti-spam minimum, which is what we want).
Verified: DNS3 stuck at snapshot height 2,214,547 with log lines
'ERROR: ProcessBlock() : block with too little proof-of-stake'
on every inbound post-snapshot block, while DNS2 (same daemon version) had advanced to
2,214,757 — confirming the issue is per-node state, not consensus.
Keeps Misbehaving(5) soft score from cbb189a (was 100, instant ban).
The writer seek calculation (contentHashPos - sizeof(numUtxos)) was
correct for v1/v2. In v3 the layout inserted numBlocks between
numUtxos and numStakeSeen, so the seek landed on the numBlocks field
and the updated count was written there, corrupting both fields.
Fix: compute the offset relative to contentHashPos, skipping the
contentHash, numStakeSeen, and numBlocks fields inserted in v2/v3.
Adds 8 new hardened checkpoints at heights 2206500-2214400, verified
against the canonical chain. Closes the 8,400-block unverified gap
between the last hardcoded checkpoint (2206004) and the live tip
(2,214,476).
Without these, a fresh node syncing from zero with NO snapshot has
zero finality protection above height 2206004. A peer feeding fork
blocks at heights 2206005-2214400 could trick the IBD node into
accepting a divergent chain, because CheckHardened() only fires at
the exact heights in mapCheckpoints.
With continuous pins every 1000 blocks, any divergence >1000 blocks
is rejected at AcceptBlock time with DoS=100, protecting from-zero
sync against low-trust forks.
Two bugs in the anti-spam heuristic at src/main.cpp:
1. Inverted comparison: condition was bnNewBlock > bnRequired paired with
"too little proof-of-stake" error message. The condition triggers when
the block has MORE difficulty than required (harder than allowed),
but the message claims the OPPOSITE. Honest blocks during legitimate
time-warps (fork recovery, chain catchup) get mislabelled.
2. Misbehaving(100) was a single-shot instant ban: banscore threshold
defaults to 100, so the FIRST anti-spam violation triggered a 24-hour
ban on every honest peer feeding us blocks during fork divergence.
This is what caused the 2026-06-23 DNS2 clearnet-fork incident:
peers got banned before we could determine which chain was canonical.
Fix: condition now correctly says bnNewBlock < bnRequired (block too
easy = reject), and Misbehaving score dropped from 100 to 5 (needs
~20 anti-spam violations before the 100 banscore threshold). Anti-spam
is a soft signal, not a hard ban trigger.
Adds v3 snapshot format that includes the last 5000 PoS block
(prevoutStake, nStakeTime) pairs so a snapshot-loaded node has its
stake-collision set restored without walking the block index.
v2 readers still load v3 snapshots (the extra field is past numBlocks
and the loader checks version >= 3 to read numStakeSeen).
Bumps version to 6.1.1.
SQLite wallet (default), RocksDB chaindb (default), Boost removal
from common link, I2P startup performance fix (180s→27ms).
Migration is automatic and non-destructive:
- Wallet: BDB → SQLite on first open (original preserved as .bdb.bak)
- ChainDB: LevelDB → RocksDB on next start (via -chaindb flag)
The cpp20 rebase removed all Boost includes from init.cpp but left
this orphaned using-directive. It compiled on Linux daemon (I2P
embedded transitively provides the boost namespace) but broke every
Qt/Windows/macOS build in CI with: 'boost' is not a namespace-name.
i2p::api::StartI2P() calls NetDb::Start() which triggers Reseed() — a
blocking HTTPS download from public I2P reseed servers. On first run
(empty netDb), this blocks for up to 180s, freezing the GUI splash
screen on "Starting embedded I2P router..." indefinitely.
Fix: Move StartI2P + client::context.Start + SOCKS/SAM readiness polling
into a detached background thread. InitI2P (fast, <1s) remains
synchronous. The main init thread proceeds immediately — Tor-only mode
works while I2P bootstraps asynchronously (1-5 min for first run).
Verified: i2p_start startup perf dropped from 180s+ to 27ms on first run
with empty netDb. Reseed + tunnel building continues in background.
Live-validation of the CWalletDB rebase (preserved snapshot wallet from
DNS2's 2026-04-23 recovery, copied to /tmp/cpp20-validate, run with the
new build) revealed that after MaybeMigrateBerkeleyWalletToSQLite()
converts wallet.dat from Berkeley DB to SQLite, the next block of init
code still calls bitdb.Verify() on the now-SQLite file. BDB cannot open
a SQLite file, so Verify returns RECOVER_FAIL, which init treats as a
fatal InitError ("wallet.dat corrupt, salvage failed"). The migration
itself succeeds and produces a valid SQLite wallet.dat + a
wallet.dat.bdb.bak copy of the original, but the daemon refuses to
continue past Step 5.
Fix: snapshot IsSQLiteFile(walletPath) AFTER the migration hook, then
gate the entire BDB-only Step 5 path (bitdb.Open, salvagewallet,
zapwallettxes, bitdb.Verify) on the snapshot. Once the wallet is SQLite
— whether by migration or because it was born that way — the BDB
verify/salvage steps are skipped; SQLiteDatabase::Open() already runs
PRAGMA integrity_check during connection setup, so the wallet is
validated against the SQLite schema before any handle is constructed.
The verify_db performance log now includes wallet_is_sqlite=N for
observability.
Validation: full migration cycle completed in the test datadir, RPC
getwalletinfo/getaddressesbyaccount returned valid data matching the
pre-migration BDB wallet exactly (5 unique HD-derived addresses,
keypoolsize=101, keypoololdest=1781772890, walletversion=60000).
Move CWalletDB off the Berkeley CDB base class and onto the typed batch
seam introduced by walletdb-batch.h / walletdb-{factory,sqlite}.{h,cpp}.
Build seam
----------
* CWalletDB now derives from CWalletBatchTyped. The typed Read/Write/
Erase/Exists templates come from the seam; their bodies (WriteTx,
WriteKey, WriteMasterKey, ReadPool, WriteSetting, ...) are unchanged
because only the base class swapped — the call signatures resolve to
the same templates.
* CWalletBatchTyped takes ownership of the WalletDatabase so the
underlying connection outlives any batch issued by it (SQLiteBatch
holds a reference, not a value). The two-phase Open() pattern lets
CWalletDB hand the freshly opened database to the base class after
MakeWalletDatabase() returns.
* MakeWalletDatabase (walletdb-factory.cpp) routes -walletdb=sqlite to
SQLiteDatabase, returning nullptr with a clear error for the
unfinished Berkeley branch. The CWalletDB constructor surfaces that
error string on failure.
Cursor sites (the only Berkeley-specific call sites)
---------------------------------------------------
Three sites used GetCursor()/ReadAtCursor() directly:
* LoadWallet — full scan, now uses StartCursor()/NextRecord()
* ListAccountCreditDebit — used DB_SET_RANGE + DB_NEXT loop; replaced
with full keyspace scan + filter-in-loop (SQLite cursor does not
support keyed range seeks). Behaviour matches Berkeley: terminates
when strType changes or, in single-account mode, when
acentry.strAccount differs.
* ZapWalletTx — moved to BerkeleyZapWalletTx (see below) because it
operates on raw Berkeley Db/Dbc/Dbt now that CWalletDB is on the
seam.
The 3 unused public methods on the old CWalletDB (GetAtCursor /
GetTxnCursor / GetAtActiveTxn) had no callers outside walletdb.{h,cpp}
(verified by grep) and were removed.
Berkeley-only escape hatches
----------------------------
Recover(CDBEnv&,...) and ZapWalletTx(...) became BerkeleyRecoverWallet
and BerkeleyZapWalletTx in a new walletdb-recover.{h,cpp} pair. They
operate directly on DbEnv/Db/Dbc/Dbt because CDB's members are
protected (free functions cannot use the wrapper). The recovery logic
duplicates a BDB-only ReadKeyValue variant locally to avoid pulling
the typed batch seam into a Berkeley-only file.
Init.cpp uses these via:
* -salvagewallet -> BerkeleyRecoverWallet(bitdb, ..., fOnlyKeys=true)
* -zapwallettxes -> BerkeleyZapWalletTx(...)
* bitdb.Verify -> BerkeleyRecoverWallet as the recover callback
Wallet migration hook
---------------------
After the Berkeley verify/salvage/zap steps and before CWalletDB is
opened for the live wallet, init.cpp now calls:
if (ResolveWalletDbKind() == SQLite &&
!IsSQLiteFile(walletPath))
MaybeMigrateBerkeleyWalletToSQLite(walletPath, err)
The migration code (walletmigrate.{h,cpp}) is unchanged — it opens a
private Berkeley environment over the wallet directory, copies every
record verbatim (raw key/value bytes) into a fresh SQLite file,
verifies the row count, then atomically renames the BDB original to
"<name>.bdb.bak" and the SQLite file into place. On any failure the
BDB original is left exactly as it was. Errors surface through
InitError so the daemon refuses to start with a corrupt wallet rather
than silently falling back to Berkeley.
No working Berkeley fallback
----------------------------
MakeWalletDatabase returns nullptr for the Berkeley branch, so
-walletdb=bdb no longer opens a working wallet through the seam. This
is intentional for this release — the migration hook handles existing
BDB wallets at first startup, after which the on-disk file is SQLite
and the BDB code path becomes pure recovery glue.
Header fallout
--------------
walletdb.h no longer pulls in db.h (which would drag <db_cxx.h> into
every TU that includes the wallet API). Forward decls added for
CWalletTx, CBlockLocator, CWallet, CPubKey, CScript, CMasterKey,
uint160, uint256. nWalletDBUpdated is now extern-declared in
walletdb.h and defined in db.cpp (was previously declared in db.h).
Validation
----------
Build: GREEN with USE_TOR_EMBEDDED=ON USE_I2P_EMBEDDED=ON. 6 binaries:
trianglesd, triangles-cli, test_triangles, test_chaindb_runtime,
test_chaindb_equivalence, test_snapshotnet.
Tests: 107/107 + 10/10 + 5/5 = byte-identical to the 5d9da84
baseline. wallet_tests and accounting_tests inside test_triangles now
exercise the SQLite path for the first time — their pass is the
de-facto wallet-migration validation at the test-suite level.
Modernizes triangles.conf parsing to use plain stdlib instead of
boost::program_options::detail::config_file_iterator.
Supports the syntax that actual triangles.conf files use:
- key=value or key = value (whitespace around = is ignored)
- # comments and blank lines are skipped
- Surrounding double quotes are stripped from values
- Same InterpretNegativeSetting semantics
- Command-line settings still take precedence (don't overwrite existing)
Intentionally NOT supported (different from Boost):
- Backslash line continuations
- Escape sequences inside quoted values
- Section headers ([section])
After this change, triangles_common no longer uses Boost.ProgramOptions
directly. The Boost.ProgramOptions link inside the USE_I2P_EMBEDDED block
remains because i2pd-src/libi2pd/Config.cpp uses it internally.
Verified:
- Build with -DUSE_I2P_EMBEDDED=ON succeeds
- Build with -DUSE_I2P_EMBEDDED=OFF succeeds
- Without i2p, ldd on trianglesd shows NO libboost_program_options dep
Merge left StopI2P() (duplicated StopEmbeddedI2P), StartI2P(),
CI2PProcess::GetInstance(), and I2P_DEFAULT_SAM_PORT references from
the SAMI-PC process-based I2P. Replaced with the v6 embedded I2P API:
- shutdown: single StopEmbeddedI2P() (was called twice plus StopI2P)
- startup: single StartEmbeddedI2P() which reads its own args
- removed manual SAM host/port resolution (StartEmbeddedI2P handles it)
Merge left residual references to the SAMI-PC process-based I2P API
(CI2PSession, fI2P) in files that now compile against the v6 embedded
I2P (CI2PEmbedded). Fixed:
- net.cpp ConnectNode: removed fI2P/CI2PSession blocks, restored
v6 SOCKS-proxy connection path (I2P routing handled in netbase)
- CMakeLists.txt: removed i2p.cpp/i2p_process.cpp from build (not
part of embedded I2P; kept in tree as reference only)
- rpcnet.cpp: CI2PSession → CI2PEmbedded (IsRunning/GetI2PAddress)
- rpcwallet.cpp: same API migration
- init.cpp: same API migration for startup address print
Merges the HD wallet work and process-based I2P integration from the
SAMI-PC hd-wallet branch into v6 master. Conflict resolution keeps
v6 embedded I2P (CI2PEmbedded) as primary, includes process-I2P
files for reference, preserves FastImportBlockFile() from hd-wallet,
and keeps v6 version numbers (6.0.0) and wAddressStack Qt layout.
i2pHostname was cleared on Start() but never populated, so
GetI2PAddress() always returned empty and the Qt status bar never
showed the I2P address even when the router was running.
Now queries i2p::context.GetRouterInfo().GetIdentHash().ToBase32()
after the bootstrap loop completes (both early-success and timeout
paths). The address appears as <hash>.b32.i2p in the status bar.
libboost_system-mt.a doesn't exist on MSYS2 (header-only in newer
Boost). find_library auto-discovers the actual filenames and skips
any that don't exist. Resolves both the missing-file error and the
naming ambiguity.
MSYS2 MinGW doesn't create CMake imported targets for Boost, so
Boost::filesystem etc. silently don't link. i2pd's own Makefile.mingw
solves this by referencing full paths like /mingw64/lib/libboost_*.a.
We do the same — auto-detect MINGW_PREFIX (/mingw64) and link the
exact .a files for boost_filesystem, boost_program_options,
boost_system, openssl, and zlib.
The linker needs to see i2pd archives, then Boost/zlib to resolve
their symbols, then i2pd archives AGAIN to resolve any remaining
references. Added raw -l fallbacks for MinGW where Boost:: CMake
imported targets may not exist even though libs are installed.
Windows: Replace --start-group/--end-group (CMake mis-orders them
with Ninja generator) with double-listing of i2pd static archives.
Linker resolves circular deps in two left-to-right passes.
macOS: Boost 1.90 via Homebrew doesn't provide filesystem/system as
separate COMPONENTS. Use OPTIONAL_COMPONENTS so find_package doesn't
fail, then guard the target_link_libraries with if(TARGET Boost::...).
libi2pd.a references boost::filesystem::detail::status, exists,
create_directories, etc. These are in boost_filesystem, which the
I2P link section was missing. Added Boost::filesystem and
Boost::system.
Windows: MinGW's rpcndr.h #defines 'interface' as 'struct' (COM
support). i2pd's I2CP.h uses it as a parameter name, causing
parse errors. Add #undef interface before i2pd includes.
macOS: i2pd Makefile.homebrew hardcodes openssl@3.5 but Homebrew
installs openssl@3. build-libi2pd.sh now detects the actual path
and passes SSLROOT=<path> to make, which overrides the Makefile
assignment.
The CF Open overload had the same DB** vs unique_ptr<DB>* API drift
as the non-CF version, but was calling rocksdb::DB::Open directly
instead of through the SFINAE wrapper. On MSYS2 MinGW (Windows CI)
the unique_ptr-only overload causes a compile error. Added
OpenRocksDBCF with the same int/long SFINAE pattern.
Replaces the single label_onion item inside the wStatusBar layout with a
vertical group (wAddressStack) containing two rows:
row 1: [I2P] <.b32.i2p address>
row 2: [Tor] <.onion address>
Both address labels now copy their text to the clipboard on click via
the existing eventFilter pattern (extended to handle labelI2PAddress in
addition to labelOnionAddress). The label_i2p, label_i2p_icon, and new
label_tor_icon widgets live in mainwindow.ui so they share the same
layout stretch and ordering as the rest of the status bar; the
QWidget/QVBoxLayout/QHBoxLayout nesting keeps the stack compact and
centered on the existing 37px -> 52px status bar height bump.
Tooltips updated to "Click to copy" for both addresses (no longer
"Selectable - right-click to copy") to match the actual behaviour.
Tooltip wording for [Tor] chip matches the existing [V3] chip.
Add Ed25519 signature support to snapshot manifests. Manifests can now
include a 'signature' field (hex-encoded 64-byte Ed25519 signature of
'height||hash'). VerifyManifest checks it against a compiled-in pubkey.
- Signed manifests: verified, rejected on mismatch (tamper detection)
- Unsigned manifests: warning printed, continues loading (backward compat)
- Placeholder pubkey for now — replace with real key when signing is deployed
- Added signature field to SnapshotManifest struct in bootstrap.h
This closes the 'loading WITHOUT signature verification' security gap
that was printed during every bootstrap download.
Add .b32.i2p address display alongside existing Tor .onion address
in the wallet status bar. Purple [I2P] indicator shows router state:
- Purple: I2P active with valid destination
- Yellow: router running, building tunnels
- Hidden: I2P not active
Updates every 5s via timer, parallel to updateOnionAddress().
BIP152 Compact Blocks (main.cpp, net.cpp, protocol.h):
- SipHash-2-4 short IDs (48-bit) for transaction identification
- Compact block relay with mempool reconstruction
- Merkle root verification before acceptance
- Graceful fallback to full block on any mismatch
- Collision detection for ambiguous short IDs
RocksDB Column Families (txdb-rocksdb.cpp/h):
- 5 CFs: default, blockindex, txindex, utxo, addrindex
- Per-CF tuning: UTXO optimized for point lookups, addrindex for scans
- Backward-compatible: falls back to default CF for pre-migration data
- Prefix-based routing in ReadRaw/WriteRaw/EraseRaw/ExistsRaw
Fork Detector (main.cpp, net.cpp, net.h):
- Background thread checks local tip vs peer median every 60s post-IBD
- Alerts on divergence > forkthreshold (default 5 blocks)
- Optional auto-rebuild trigger on severe divergence
Cross-Network Tor↔I2P Discovery (net.cpp, init.cpp):
- I2P seed addresses loaded into addrman alongside onion seeds
- Address relay bridges .onion and .b32.i2p between networks
- IsI2PAddr/IsOnionAddr helpers for network-type detection
Configurable Outbound Connections (net.cpp, init.cpp):
- -maxoutboundconnections flag (range 4-32, default 8)
Mempool Fee-Priority Boost (miner.cpp):
- 2x fee weight in PoS block assembly for higher staking rewards
SAM v3 Direct Streaming (i2p/i2p_embedded.cpp/h):
- CI2PSamSocket class with full SAM v3 protocol
- SESSION CREATE + STREAM CONNECT handshake
- Factory method on CI2PEmbedded for native I2P connections
- SAM bridge readiness check in bootstrap loop
When smsgDB is opened by a binary linked against an older RocksDB than
the one that wrote its SST files, Open() returns
'Corruption: unknown checksum type 4 in .../000064.sst ...' (XXH3 was
introduced in RocksDB 7.4). Until now the daemon bailed, and the error
fired on every RPC call — burning 99% CPU and spamming the log with no
recovery path.
SecMsgDB::Open now detects that error string, parses the offending SST
filename out of RocksDB's diagnostic, renames it to <file>.sst.quarantined-<unix-ts>
inside smsgDB/, and retries the open. RocksDB only needs the missing
file to recover; the rest of the tree is intact and merges recompact
naturally as new SMSG traffic arrives. Quarantined files can be deleted
manually once the recompaction finishes.
CMakeLists.txt now refuses to configure against RocksDB < 7.4.0 when
the version is detectable (find_package or pkg-config paths). The
manual-probe path (Ubuntu 22.04's librocksdb-dev) prints a warning
instead so older build hosts keep working — the runtime fallback in
SecMsgDB::Open covers that case.
Discovered 2026-06-27 on DNS2: a Jun 19 binary swap left
smsgDB/000064.sst written with XXH3; the current v5.9.24 daemon is
linked to librocksdb.so.6.11 (RocksDB 6.11) which can't read it.
Behaviour before this patch: 99% CPU, log spam on every RPC.
Behaviour after: one quarantine log line, daemon proceeds normally.
Refs: the existing pre-v5.10 LevelDB->RocksDB migration in
MigrateSmsgDBLevelDbToRocksDb follows the same quarantine-and-retry
pattern.
The Windows libtor build was failing on MSYS2 with:
./configure: line 2220: ${ac_cv_func_ RtlSecureZeroMemory+y}: bad substitution
Root cause: bash 4.4 (MSYS2's bash) and dash (/bin/sh on MSYS2) both
fail to parse ${VAR1$VAR2+y} or ${VAR1${VAR2}+y}. autoconf 2.69-2.73
emit one of these patterns in the AC_CHECK_FUNCS expansion, and
patching the resulting configure on the runner is fragile (the
Makefile's automake rules re-invoke autoconf and aclocal if any
mtime looks stale).
Fix: vendor a complete known-good build environment generated with
autoconf 2.71 on Linux. The vendored set:
src/tor/configure.vendored (37,966 lines, bash 4.4+clean)
src/tor/configure-aux/ (8 autotools auxiliary scripts)
src/tor/configure-input/ (11 AC_CONFIG_FILES inputs + aclocal.m4)
src/tor/regenerate-tor-configure.sh (one-shot regenerator with parse check)
src/tor/build-libtor.sh (uses vendored set when present)
build-libtor.sh now:
1. Copies configure.vendored + 8 aux files + 11 inputs into the
tor-src submodule directory.
2. Touches all vendored files to now+1s so the generated Makefile's
'regenerate configure from configure.ac' and 'regenerate
aclocal.m4 from m4/' rules see no work to do.
3. Runs configure directly (skips autoreconf entirely).
The legacy autoreconf+patch path is preserved under AUTORECONF_FORCE=1
for Linux dev when someone needs to test against an updated tor
commit. regenerate-tor-configure.sh handles regenerating the
vendored set from a fresh autoconf run.
Workflow:
build-all.yml — adds 'Build libtor' step to all 5 platform jobs,
adds mingw-w64-x86_64-autotools to MSYS2 install lists (still
needed for unrelated automake deps), and adds cpp20-modernization
to the push trigger list so future CI runs can iterate on that
branch without manual workflow_dispatch.
Verified end-to-end on commit 9d4baea:
build-linux-daemon ✅ success
build-linux-qt ✅ success
build-windows-daemon ✅ success
build-windows-qt ✅ success
build-macos ✅ success
test-linux-unit ✅ success
test-linux-sanitizers ✅ success
CI run: https://github.com/SamiAhmed7777/triangles_v5/actions/runs/28209275346
- txdb-factory.cpp: drop static-cache in ResolveChainDbKind so the
-chaindb flag can be toggled at runtime (needed for tests; cost is
negligible since the daemon sets it once at startup)
- txdb-rocksdb.cpp: fix ExistsRaw to honor pending-batch delete markers.
Previously a key erased inside an open batch was still reported as
existing because the underlying DB hadn't been updated yet. Mirror
ReadRaw's correct behavior: a delete marker shadows the DB value.
- chaindb_runtime_tests.cpp: per-test fresh handle via close-reopen
dance so the static g_rocksdb singleton doesn't leak state between
cases. Tests filter framework keys (length-prefixed 'version' and
'dbformat') from iterator walks. block_index test fixed to Seek()
not Seek("blockindex") since the serialized keys start with the
length byte 0x0a.
- snapshotnet_tests.cpp, chaindb_runtime_tests.cpp: include wallet.h,
ui_interface.h, uint256.h, checkpoints.h as needed for linker; add
BOOST_TEST_MODULE decl; define global stubs (pwalletMain,
uiInterface, fConfChange, etc.) so wallet.cpp link succeeds.
Result: test_snapshotnet + test_chaindb_runtime both pass with zero
errors. Found and fixed a real production bug in ExistsRaw along
the way.
- snapshotnet.cpp: always re-scan on HasServableSnapshot; auto-dump
from current chain when synced to canonical snapshot height
- net.cpp: EnsureLocalSnapshot() at startup so NODE_SNAPSHOT reaches
outbound peers in the first version message
- CMakeLists.txt: add test_snapshotnet + test_chaindb_runtime targets
- test/snapshotnet_tests.cpp, test/chaindb_runtime_tests.cpp: full
coverage for the SnapshotNet P2P protocol + CRocksTxDB wrapper layer
Run #473 (post CONFIG_SHELL=bash) still hit:
./configure: line 11244: syntax error near unexpected token
`as_ac_var=`printf '%s\n' "ac_cv_func_$ac_func" | sed "$as_sed_sh"``
Root cause: MSYS2's mingw-w64-x86_64-autotools meta package pulls
autoconf 2.73, which generates ./configure with backtick command
substitution INSIDE variable assignments (`var=`cmd``). My local
environment has autoconf 2.71 which doesn't generate this pattern
at all (verified: 0 matches in locally-generated configure).
bash on MSYS2's MINGW64 can't parse the 2.73 pattern even when
invoked directly - the nested backticks with mixed single/double
quotes containing $-vars trip the parser. Pinning MSYS2's autoconf
to 2.71 is fragile (meta-package pulls current on next rebuild).
Fix: after autoreconf, run a perl one-liner on the generated
configure that converts all `var=`cmd`` assignments to
`var=$(cmd)` form. POSIX-ly equivalent for bash, nests cleanly,
and matches what autoconf 2.71 would have generated. Verified
the patched configure still works (`./configure --help` runs
cleanly). The CONFIG_SHELL=bash line stays for any remaining
edge cases on dash-vs-bash differences.
Krystie
sami7777