dash: emit the canonical p2pool coinbase marker so explorers attribute our blocks (+ wire --coinbase-text) by frstrtr · Pull Request #901 · frstrtr/c2pool

frstrtr · 2026-07-26T14:17:40Z

Problem

Block explorers attribute blocks to a pool by coinbase text.
chainz.cryptoid.info/dash/extraction.dws?30.htm
registers this pool as P2Pool-DASH and has no knowledge of the string c2pool.

c2pool was writing scriptSig 03c751266332706f6f6c = BIP34 height push + bare c2pool
(verified on our own live mainnet block 2511303). We are ~87% of the p2pool-dash pool's
hashrate and the two blocks won on 2026-07-24 (2511241, 2511303) are credited to nobody.

Second, smaller problem found on the way: --coinbase-text is an existing, documented
feature (README.md "Coinbase structure", src/core/coinbase_builder.hpp) that
main_ltc.cpp wires and main_dash.cpp never did — it did not appear in the DASH
binary's --help at all, and the README claimed a /c2pool/ default that the DASH lane
did not even use.

What this changes (DASH only)

Default coinbase text, sourced from the coin SSOT:

network	scriptSig	rendered
mainnet (h=2511303)	`03c75126` `2f5032506f6f6c2d444153482f6332706f6f6c2f`	push3 height + `/P2Pool-DASH/c2pool/` (24 B)
testnet / regtest (h=950000)	`03f07e0e` `2f5032506f6f6c2d74444153482f6332706f6f6c2f`	push3 height + `/P2Pool-tDASH/c2pool/` (25 B)

src/impl/dash/config_pool.hpp now carries COINBASEEXT_HEX /
TESTNET_COINBASEEXT_HEX transcribed verbatim from the oracle
(p2pool-dash networks/dash.py:11, dash_testnet.py:11), push opcode included, so the
SSOT pins the oracle constant exactly. coinbaseext_text() strips that opcode; what we
emit is plain text, because the slot is operator-overridable and a bare 0x0D control
byte would vanish the moment anyone set --coinbase-text. Attribution is unaffected —
explorers key on the ASCII, and /P2Pool-DASH/ is byte-identical to what a canonical
p2pool-dash node renders.

--coinbase-text wired into main_dash.cpp, mirroring the LTC pattern: parsed,
bounded by c2pool::MAX_OPERATOR_TEXT_SOLO (64 B — DASH has no merged-mining commitment
sharing the 100-byte budget), resolved once before any coinbase is built, echoed at
startup, and present in --help.

One scriptSig SSOT. DASH built the scriptSig in two places — coinbase::build()
(stratum job / block coinbase) and mint::build_producer_job() (share_data['coinbase']).
A one-byte divergence between them makes the node self-reject its own shares. Both now
call dash::coinbase::build_coinbase_scriptsig(), and a KAT pins their equality. This is
also why the override lives on the SSOT rather than being threaded per-call.

Consensus safety

Verified before touching bytes, not assumed:

Not re-derived by peers. The scriptSig travels on the wire as
share_info.share_data.coinbase (VarStrType, 2 <= len <= 100, data.py:315).
Share.check() calls generate_transaction(tracker, SELF.share_info['share_data'], ...)
— it feeds the received share's own coinbase field back in, so the re-derived gentx
contains whatever bytes the sender wrote. COINBASEEXT appears nowhere in the
oracle's data.py; grep puts it only in networks/*.py and the stratum assembly at
work.py:339. c2pool's mirror share_check.hpp::generate_share_transaction likewise
does tx << share.m_coinbase.
share_info != self.share_info is satisfied for the same reason — share_data is
passed through, not rebuilt.
hash_link is length-agnostic. prefix_to_hash_link() takes a SHA-256 midstate over
the gentx prefix plus extra_data = trailing partial 64-byte block minus the
const_ending; check_hash_link() reconstructs it from hash_link.length % 64. Both
sides derive the split from the actual prefix length, and dash's hash_link_type
carries extra_data as a VarStr precisely so a variable-length prefix works (the
upstream assert not extra_data is commented out in the oracle for this reason).
Empirically decisive. c2pool has been writing c2pool — matching no network
constant — and canonical p2pool-dash peers accept its shares with zero bans across
thousands of shares and two dashd-confirmed mainnet blocks.

Framing (not consensus, but breaking it breaks mining)

BIP34 height push stays first and is never reached by the cap. dashd's
ContextualCheckBlock compares the scriptSig prefix to CScript() << nHeight.
Worst-case push is 5 bytes; --coinbase-text is capped at 64; 5 + 64 = 69 < 100.
Tested with a 400-byte text: result is exactly 100 bytes with the height push intact.
Extranonce offsets do not move. DASH's stratum extranonce2 is the 8-byte nonce64
inside the OP_RETURN output, not the scriptSig. coinb1/coinb2 split at
nonce64_offset, which build() derives from the tx tail
(total - payload - 4 - 8). A longer scriptSig lengthens coinb1 only; the advertised
extranonce2_size (8) and coinb2 are unchanged. Asserted in the KAT.

Other lanes — reported, deliberately not changed

lane	emitted today	source
LTC	`/c2pool/`	hardcoded twice in `src/core/web_server.cpp` (756, 1103) + twice in `merged/merged_mining.cpp`
BTC	`/c2pool-btc/`	`src/impl/btc/stratum/work_source.cpp:430`
BCH	`/c2pool-bch/`	`src/impl/bch/stratum/work_source.cpp:415`
DGB	nothing — empty scriptSig, no BIP34 height either	`main_dgb.cpp:861-868` never sets `ain.coinbase_script`
DASH	`c2pool` -> `/P2Pool-DASH/c2pool/`	this PR

Adjacent findings, filed here rather than fixed:

DGB emits an empty coinbase scriptSig on the stratum path — no pool tag and no
BIP34 height push. That looks like a real bug, not a cosmetic gap.
BTC/BCH writers emit tags their own readers cannot match (block_json.hpp:99 and
bch/coinbase_commitment.hpp:198 both look for /c2pool/).
Only p2pool-dash defines a COINBASEEXT among the oracles available on this box;
jtoomim's networks/bitcoin.py has none (checked directly). No p2pool-ltc / -dgb / -bch
checkout exists locally, so their canonical markers are unverified.

Tests

Folded into the existing allowlisted test_dash_coinbase_parity target — no new
add_executable. 9 new cases; per #895, each was confirmed to actually execute (gtest
[ RUN ]/[ OK ] lines, not a bare ctest tick):

[  PASSED  ] 19 tests.   (10 pre-existing + 9 new)
  DashCoinbaseMarker.MainnetScriptSigKAT
  DashCoinbaseMarker.TestnetScriptSigKAT
  DashCoinbaseMarker.LegacyBareTagIsNotAttributable        <- negative control
  DashCoinbaseMarker.ConfigPoolCarriesOracleConstant
  DashCoinbaseMarker.OperatorOverrideReplacesDefault
  DashCoinbaseMarker.TruncationRespectsBoundAndKeepsHeightPush
  DashCoinbaseMarker.BuildEmitsMarkerInSerializedCoinbase
  DashCoinbaseMarker.BuildHonoursTestnetParams
  DashCoinbaseMarker.MintAndBlockCoinbaseScriptSigsAgree

Negative control run: flipping one nibble of COINBASEEXT_HEX
(...444153482F -> ...444153483F) and rebuilding fails 5 of the 9 with the expected
byte diff ("...444153483f..." vs "...444153482f...") and find(...) == npos on the
marker — so the KATs discriminate rather than tautologically pass. Reverted.

Regression sweep: all 41 test_dash_* binaries built and run — 0 failures.
test_dash_stratum_work_source.cpp:389 had hardcoded "c2pool" in its
stratum-vs-SSOT byte-identity comparison and was updated to read the SSOT (pinning the
literal there would break the test the moment an operator sets --coinbase-text, which is
the whole point of the flag).

--help on the built binary:

       c2pool-dash --run ...
           [--message-blob-hex HEX] [--coinbase-text TEXT]
...
        --coinbase-text TEXT sets the coinbase scriptSig text written
        after the BIP34 height push (README "Coinbase structure"; max
        64 bytes, no merged mining on the DASH lane). Default
        "/P2Pool-DASH/c2pool/" (testnet "/P2Pool-tDASH/c2pool/") --
        the /P2Pool-DASH/ marker is what block explorers match on to
        attribute a block to this pool; the c2pool suffix says which
        implementation mined it. ...

Bound enforced: --coinbase-text with 65 bytes prints
[args] --coinbase-text too long: 65 bytes (max 64) and exits 2; 64 bytes is accepted.

Could not verify / deliberately out of scope

regtest gets the tDASH string, not an rDASH one. main_dash.cpp maps --regtest
onto testnet = true — DASH has no third sharechain profile — so a separate
/P2Pool-rDASH/ would need a three-way network selector across the whole lane. Noted in
the SSOT comment. Also: the oracle's own dash_regtest.py:11 constant
0F2F5032506F6F6C2D724441534828 is malformed — it declares push-15 but carries 14
bytes, and ends 0x28 ( where / was clearly meant (/P2Pool-rDASH(). Not
reproduced. (dash_custom_example.py:25 has the same class of bug: push-13 declared,
10 bytes present.)
No live-explorer confirmation. The cryptoid extraction page currently renders
"Coming soon... :)" for its block table, so I could confirm the pool is registered as
P2Pool-DASH but not observe its matching rule directly. The argument for
/P2Pool-DASH/ is that it is byte-identical to what canonical p2pool-dash nodes write.
Canonical markers for LTC/DGB/BCH are unverified — no local p2pool checkouts for
those chains.
pool.yaml coinbase_text is not wired on DASH — main_dash.cpp has no YAML config
loader at all (0 hits for cfg[/Fileconfig). CLI only. The README table row keeps the
YAML key because it is accurate for LTC.
No soak. Nothing here was exercised against a live sharechain or a real dashd; the
evidence is KATs plus the structural argument above.

Conflict surface

main_dash.cpp help-text block (~line 245) overlaps dash: wire the embedded arm so --run can actually take it (#738) #891's @@ -238,6 +239,15 @@ hunk —
trivial textual conflict, adjacent additions.
test/test_dash_stratum_work_source.cpp edit is at ~389; dash: wire the embedded arm so --run can actually take it (#738) #891 appends at 1646+. No overlap.
src/impl/dash/stratum/work_source.cpp edit is at 695; coins: mint the block-winning share onto the sharechain (dash/btc/bch/dgb) #888's dash hunks start at 816
(mining_submit). No overlap.

DRAFT — do not merge.

Block explorers attribute blocks to a pool by coinbase text. chainz.cryptoid.info/dash/extraction.dws?30.htm registers this pool as "P2Pool-DASH" and has no knowledge of the string "c2pool", so the blocks c2pool won for the p2pool-dash sharechain (2511241, 2511303) were credited to nobody. c2pool was writing scriptSig `03c751266332706f6f6c` = BIP34 height push + bare "c2pool". DASH now defaults to the oracle COINBASEEXT payload plus an implementation tag: mainnet 03c75126 2f5032506f6f6c2d444153482f6332706f6f6c2f = push3 height || "/P2Pool-DASH/c2pool/" (24 bytes) testnet 03f07e0e 2f5032506f6f6c2d74444153482f6332706f6f6c2f = push3 height || "/P2Pool-tDASH/c2pool/" (25 bytes) Sourced from the coin SSOT (config_pool.hpp COINBASEEXT_HEX, transcribed verbatim from p2pool-dash networks/dash.py:11 / dash_testnet.py:11 with the push opcode; coinbaseext_text() strips that opcode so the emitted value is plain text an operator can override). --coinbase-text was a documented, implemented feature (README "Coinbase structure", src/core/coinbase_builder.hpp) that main_ltc.cpp wires and main_dash.cpp never did: it was absent from the DASH binary's --help and unusable on that lane. It is now parsed, bounded by c2pool::MAX_OPERATOR_TEXT_SOLO (64 B; DASH has no merged-mining commitment sharing the budget), resolved once before any coinbase is built, and echoed at startup. The override lives on the coin SSOT rather than being threaded through each call because DASH builds the scriptSig in TWO places -- coinbase::build() for the stratum job / block coinbase and mint::build_producer_job() for share_data['coinbase']. A one-byte divergence between them would make the node self-reject its own shares. Both now go through dash::coinbase::build_coinbase_scriptsig(), and a KAT pins their equality. Not consensus-bearing. COINBASEEXT appears nowhere in the oracle's data.py; the scriptSig travels as share_info.share_data.coinbase (VarStr, 2..100 B) and Share.check() re-derives the gentx from the RECEIVED share's own coinbase field. Framing is preserved and tested: the BIP34 height push stays first and cannot be reached by the 100-byte cap (worst-case 5-byte push + 64-byte operator max), and the stratum extranonce2 slot does not move -- DASH's extranonce2 is the 8-byte nonce64 in the OP_RETURN, and coinb1/coinb2 split at an offset derived from the tx tail. Other lanes are unchanged and reported, not touched: LTC writes /c2pool/, BTC /c2pool-btc/, BCH /c2pool-bch/, DGB writes an empty scriptSig on the stratum path. Only p2pool-dash defines a COINBASEEXT among the oracles available here. README updated: the config table no longer claims a single /c2pool/ default, and a per-lane operator-text table plus the attribution rationale were added. Tests fold into the existing allowlisted test_dash_coinbase_parity target (9 new cases, no new add_executable): mainnet + testnet scriptSig KATs, the pre-change bytes as a negative control, SSOT/oracle-constant pins, the --coinbase-text override, the 100-byte bound with BIP34 intact, end-to-end bytes inside the serialized coinbase with the coinb1/coinb2 split checked, and mint-vs-block scriptSig equality.

frstrtr mentioned this pull request Jul 26, 2026

dgb: production coinbase scriptSig has no pool tag and unproven BIP34 — masked by #884, will bite when the mint seam is bound #902

Open

frstrtr marked this pull request as ready for review July 26, 2026 17:08

frstrtr merged commit 6e97849 into master Jul 26, 2026
28 checks passed

This was referenced Jul 26, 2026

build: test source lists serialize every PR on one line — split to one-source-per-line so concurrent PRs stop conflicting #908

Open

build(windows): c2pool-dash fails C1128 — main_dash.cpp exceeds MSVC object section limit, needs /bigobj #909

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

dash: emit the canonical p2pool coinbase marker so explorers attribute our blocks (+ wire --coinbase-text) - #901

dash: emit the canonical p2pool coinbase marker so explorers attribute our blocks (+ wire --coinbase-text)#901
frstrtr merged 1 commit into
masterfrom
dash/coinbase-p2pool-marker

frstrtr commented Jul 26, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

frstrtr commented Jul 26, 2026

Problem

What this changes (DASH only)

Consensus safety

Framing (not consensus, but breaking it breaks mining)

Other lanes — reported, deliberately not changed

Tests

Could not verify / deliberately out of scope

Conflict surface

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant