Ask a value what it is before asking what it holds - #765
Merged
Conversation
![sourcery-ai[bot]](https://avatars.githubusercontent.com/in/48477?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Sorry @fametrano, you have reached your weekly rate limit of 500000 diff characters.
Please try again later or upgrade to continue using Sourcery
Contributor
Reviewer's GuideThis PR adds explicit type validation and consistent BTClibTypeError/BTClibValueError handling around ten call sites that previously relied on bare Python TypeError/AttributeError, and factors BlockHeader type checks into a helper, with tests and changelog updated to match the new integer/number policy. Sequence diagram for merkle_proof verification using merkle_root_from_branchsequenceDiagram
actor Caller
participant merkle_proof
participant hashes
Caller->>merkle_proof: verify(index, branch, root)
merkle_proof->>hashes: merkle_root_from_branch(index, branch, root)
alt [index not is_integer]
hashes-->>Caller: BTClibTypeError
else [index is_integer]
hashes-->>merkle_proof: computed_root
merkle_proof-->>Caller: verification_result
end
File-Level Changes
Tips and commandsInteracting with Sourcery
Customizing Your ExperienceAccess your dashboard to:
Getting Help
|
fametrano
force-pushed
the
sig-hash-widths
branch
from
August 13, 2026 17:48
a72ea71 to
a86303b
Compare
fametrano
force-pushed
the
block-type-guards
branch
from
August 13, 2026 17:49
1a43513 to
c1316b7
Compare
fametrano
force-pushed
the
sig-hash-widths
branch
from
August 13, 2026 19:33
a86303b to
d2f8daf
Compare
Ten places in `block/`, `hashes` and `utils` compared, added to or read an attribute off an argument nothing had checked. `"5" <= 16`, `"2015" + 1` and `"hard" <= 0` are bare TypeErrors about operands, raised from underneath the library and naming neither the parameter nor the function; `.tzinfo` on a str is an AttributeError, outside *both* halves of this library's exception contract, so nothing a caller is told to catch would have caught it; and two datetimes given as unix timestamps subtract to an int, whose missing `total_seconds` is an AttributeError again. The guard is `var_int.serialize`'s and the vocabulary `is_integer`'s: a bool is not a number, it being the height one, the block count one, the leaf index one. `bip34_commitment`, and `Block.assert_valid_coinbase_height` through it; `BlockHeader.assert_valid`'s timestamp and `assert_valid_time`'s `now`; `mining.mine`'s max_tries and its header, which `dataclasses.replace` used to complain about; `next_bits`'s two datetimes; `retarget_first_height`; all three numbers of `hash_rate`, where a difficulty and a timespan are float and an integer is one of those; `hashes.merkle_root_from_branch`'s leaf index, which is what `merkle_proof.assert_as_valid` and `merkle_proof.verify` reach it through; and `utils.encode_num`. `BlockHeader._assert_valid_types` is an extraction and not only a check: `assert_valid` was one branch under C901's ten. The bool half of the eight new integer guards goes in `tests/integer_policy_test.py`, where the other twenty-five already are. The fourth slice of #744.
fametrano
force-pushed
the
block-type-guards
branch
from
August 13, 2026 19:34
c1316b7 to
ab649e0
Compare
It is the coercion every `Octets` parameter of the library runs through, and it had two holes. A hex string that is not one left through `bytes.fromhex`'s bare ValueError -- the class the contract promises, with nothing saying it came from here. And anything that was not a `str` went through *untouched*, so `len` of a tuple of 33 ints was 33, which is how `taproot.assert_valid_control_block` accepted one as a control block size and how `bin_str_entropy_from_entropy(())` was reported as zero bits. Both are refused now. Every buffer is still taken, and returned as it came: a read must not rewrite the field it reads, which is what `bytes()` here would do to a bytearray a caller built. The message is `bytes.fromhex`'s own, which names a position and never the string: an Octets parameter is candidate key material as often as not, and `to_prv_key` puts this very message inside its own "not a private key" (issue #137). `to_prv_key`'s two "it must be octets" fallbacks catch a TypeError beside the ValueError, as `to_pub_key` already did. `mnemonic.entropy`'s five `int(x, 2)` parses and its `int(x, 16)` and `int(x)` ones are BTClibValueError too, and say a base rather than the digits. `borromean.sign` checks that its rings, signing indexes and nonces are of one length rather than leaving it to `zip(strict=True)`, whose message named "argument 3" and no parameter of the function; strict=True stays, as the assertion that the check and the loops cannot drift apart. Six test files stop asserting the bare class, `docs/source/guide.rst` shows the new spelling, and HISTORY.md carries the breaking-changes bullet for this and for the three slices before it. The last slice of #744.
fametrano
added a commit
that referenced
this pull request
Aug 13, 2026
* Ask a value what it is before asking what it holds Ten places in `block/`, `hashes` and `utils` compared, added to or read an attribute off an argument nothing had checked. `"5" <= 16`, `"2015" + 1` and `"hard" <= 0` are bare TypeErrors about operands, raised from underneath the library and naming neither the parameter nor the function; `.tzinfo` on a str is an AttributeError, outside *both* halves of this library's exception contract, so nothing a caller is told to catch would have caught it; and two datetimes given as unix timestamps subtract to an int, whose missing `total_seconds` is an AttributeError again. The guard is `var_int.serialize`'s and the vocabulary `is_integer`'s: a bool is not a number, it being the height one, the block count one, the leaf index one. `bip34_commitment`, and `Block.assert_valid_coinbase_height` through it; `BlockHeader.assert_valid`'s timestamp and `assert_valid_time`'s `now`; `mining.mine`'s max_tries and its header, which `dataclasses.replace` used to complain about; `next_bits`'s two datetimes; `retarget_first_height`; all three numbers of `hash_rate`, where a difficulty and a timespan are float and an integer is one of those; `hashes.merkle_root_from_branch`'s leaf index, which is what `merkle_proof.assert_as_valid` and `merkle_proof.verify` reach it through; and `utils.encode_num`. `BlockHeader._assert_valid_types` is an extraction and not only a check: `assert_valid` was one branch under C901's ten. The bool half of the eight new integer guards goes in `tests/integer_policy_test.py`, where the other twenty-five already are. The fourth slice of #744. * Make `bytes_from_octets` the whole of what Octets means (#766) It is the coercion every `Octets` parameter of the library runs through, and it had two holes. A hex string that is not one left through `bytes.fromhex`'s bare ValueError -- the class the contract promises, with nothing saying it came from here. And anything that was not a `str` went through *untouched*, so `len` of a tuple of 33 ints was 33, which is how `taproot.assert_valid_control_block` accepted one as a control block size and how `bin_str_entropy_from_entropy(())` was reported as zero bits. Both are refused now. Every buffer is still taken, and returned as it came: a read must not rewrite the field it reads, which is what `bytes()` here would do to a bytearray a caller built. The message is `bytes.fromhex`'s own, which names a position and never the string: an Octets parameter is candidate key material as often as not, and `to_prv_key` puts this very message inside its own "not a private key" (issue #137). `to_prv_key`'s two "it must be octets" fallbacks catch a TypeError beside the ValueError, as `to_pub_key` already did. `mnemonic.entropy`'s five `int(x, 2)` parses and its `int(x, 16)` and `int(x)` ones are BTClibValueError too, and say a base rather than the digits. `borromean.sign` checks that its rings, signing indexes and nonces are of one length rather than leaving it to `zip(strict=True)`, whose message named "argument 3" and no parameter of the function; strict=True stays, as the assertion that the check and the loops cannot drift apart. Six test files stop asserting the bare class, `docs/source/guide.rst` shows the new spelling, and HISTORY.md carries the breaking-changes bullet for this and for the three slices before it. The last slice of #744.
fametrano
added a commit
that referenced
this pull request
Aug 13, 2026
* Ask a value what it is before asking what it holds Ten places in `block/`, `hashes` and `utils` compared, added to or read an attribute off an argument nothing had checked. `"5" <= 16`, `"2015" + 1` and `"hard" <= 0` are bare TypeErrors about operands, raised from underneath the library and naming neither the parameter nor the function; `.tzinfo` on a str is an AttributeError, outside *both* halves of this library's exception contract, so nothing a caller is told to catch would have caught it; and two datetimes given as unix timestamps subtract to an int, whose missing `total_seconds` is an AttributeError again. The guard is `var_int.serialize`'s and the vocabulary `is_integer`'s: a bool is not a number, it being the height one, the block count one, the leaf index one. `bip34_commitment`, and `Block.assert_valid_coinbase_height` through it; `BlockHeader.assert_valid`'s timestamp and `assert_valid_time`'s `now`; `mining.mine`'s max_tries and its header, which `dataclasses.replace` used to complain about; `next_bits`'s two datetimes; `retarget_first_height`; all three numbers of `hash_rate`, where a difficulty and a timespan are float and an integer is one of those; `hashes.merkle_root_from_branch`'s leaf index, which is what `merkle_proof.assert_as_valid` and `merkle_proof.verify` reach it through; and `utils.encode_num`. `BlockHeader._assert_valid_types` is an extraction and not only a check: `assert_valid` was one branch under C901's ten. The bool half of the eight new integer guards goes in `tests/integer_policy_test.py`, where the other twenty-five already are. The fourth slice of #744. * Make `bytes_from_octets` the whole of what Octets means (#766) It is the coercion every `Octets` parameter of the library runs through, and it had two holes. A hex string that is not one left through `bytes.fromhex`'s bare ValueError -- the class the contract promises, with nothing saying it came from here. And anything that was not a `str` went through *untouched*, so `len` of a tuple of 33 ints was 33, which is how `taproot.assert_valid_control_block` accepted one as a control block size and how `bin_str_entropy_from_entropy(())` was reported as zero bits. Both are refused now. Every buffer is still taken, and returned as it came: a read must not rewrite the field it reads, which is what `bytes()` here would do to a bytearray a caller built. The message is `bytes.fromhex`'s own, which names a position and never the string: an Octets parameter is candidate key material as often as not, and `to_prv_key` puts this very message inside its own "not a private key" (issue #137). `to_prv_key`'s two "it must be octets" fallbacks catch a TypeError beside the ValueError, as `to_pub_key` already did. `mnemonic.entropy`'s five `int(x, 2)` parses and its `int(x, 16)` and `int(x)` ones are BTClibValueError too, and say a base rather than the digits. `borromean.sign` checks that its rings, signing indexes and nonces are of one length rather than leaving it to `zip(strict=True)`, whose message named "argument 3" and no parameter of the function; strict=True stays, as the assertion that the check and the loops cannot drift apart. Six test files stop asserting the bare class, `docs/source/guide.rst` shows the new spelling, and HISTORY.md carries the breaking-changes bullet for this and for the three slices before it. The last slice of #744.
fametrano
added a commit
that referenced
this pull request
Aug 13, 2026
* Check every width a sig_hash preimage writes, and every index `int.to_bytes` answers a field too wide for it with an OverflowError, an ArithmeticError that no `except BTClibValueError` catches; a list index out of range is an IndexError, a LookupError, outside it too. #724 made the version and lock-time checks unconditional in `Tx.serialize`, which closed those two fields for `legacy` -- the one sig_hash routed through it. `segwit_v0` and `taproot` assemble their preimage from their own `to_bytes` calls and never reach it, and `TxIn.serialize` and `TxOut.serialize` check nothing when told not to, so the same leak survived on the sequence, on the output value and on the outpoint's vout. The checks go in the serializations rather than in the callers -- `_serialized_4_byte_field`, `_serialized_camount`, `_serialized_out_point`, `_serialized_output`, `_serialized_spend_type` -- so `PrecomputedTxData` and both `sha_`/`hash_` families get them for free. `legacy` checks what its branches leave behind, NONE dropping the outputs it does not commit to. `vin_i` is bounded in `legacy`, `segwit_v0`, `from_tx` and `taproot_annex_and_ext` as it is in `taproot`; `from_tx` also refuses a prevout list of a different length from the vin, with the message `PrecomputedTxData` gives for the same mismatch. `ext_flag` is seven bits, and `message_extension` takes Octets like every other octets parameter here. `_assert_valid_4_byte_field` is imported from `tx.tx` rather than written a second time, and `_serialized_spend_type` is an extraction and not only a check: `taproot` was one branch under C901's ten. The third slice of #744. * Ask a value what it is before asking what it holds (#765) * Ask a value what it is before asking what it holds Ten places in `block/`, `hashes` and `utils` compared, added to or read an attribute off an argument nothing had checked. `"5" <= 16`, `"2015" + 1` and `"hard" <= 0` are bare TypeErrors about operands, raised from underneath the library and naming neither the parameter nor the function; `.tzinfo` on a str is an AttributeError, outside *both* halves of this library's exception contract, so nothing a caller is told to catch would have caught it; and two datetimes given as unix timestamps subtract to an int, whose missing `total_seconds` is an AttributeError again. The guard is `var_int.serialize`'s and the vocabulary `is_integer`'s: a bool is not a number, it being the height one, the block count one, the leaf index one. `bip34_commitment`, and `Block.assert_valid_coinbase_height` through it; `BlockHeader.assert_valid`'s timestamp and `assert_valid_time`'s `now`; `mining.mine`'s max_tries and its header, which `dataclasses.replace` used to complain about; `next_bits`'s two datetimes; `retarget_first_height`; all three numbers of `hash_rate`, where a difficulty and a timespan are float and an integer is one of those; `hashes.merkle_root_from_branch`'s leaf index, which is what `merkle_proof.assert_as_valid` and `merkle_proof.verify` reach it through; and `utils.encode_num`. `BlockHeader._assert_valid_types` is an extraction and not only a check: `assert_valid` was one branch under C901's ten. The bool half of the eight new integer guards goes in `tests/integer_policy_test.py`, where the other twenty-five already are. The fourth slice of #744. * Make `bytes_from_octets` the whole of what Octets means (#766) It is the coercion every `Octets` parameter of the library runs through, and it had two holes. A hex string that is not one left through `bytes.fromhex`'s bare ValueError -- the class the contract promises, with nothing saying it came from here. And anything that was not a `str` went through *untouched*, so `len` of a tuple of 33 ints was 33, which is how `taproot.assert_valid_control_block` accepted one as a control block size and how `bin_str_entropy_from_entropy(())` was reported as zero bits. Both are refused now. Every buffer is still taken, and returned as it came: a read must not rewrite the field it reads, which is what `bytes()` here would do to a bytearray a caller built. The message is `bytes.fromhex`'s own, which names a position and never the string: an Octets parameter is candidate key material as often as not, and `to_prv_key` puts this very message inside its own "not a private key" (issue #137). `to_prv_key`'s two "it must be octets" fallbacks catch a TypeError beside the ValueError, as `to_pub_key` already did. `mnemonic.entropy`'s five `int(x, 2)` parses and its `int(x, 16)` and `int(x)` ones are BTClibValueError too, and say a base rather than the digits. `borromean.sign` checks that its rings, signing indexes and nonces are of one length rather than leaving it to `zip(strict=True)`, whose message named "argument 3" and no parameter of the function; strict=True stays, as the assertion that the check and the loops cannot drift apart. Six test files stop asserting the bare class, `docs/source/guide.rst` shows the new spelling, and HISTORY.md carries the breaking-changes bullet for this and for the three slices before it. The last slice of #744. * Gate the input-validation rule, and let the gate enumerate (#774) `tests/input_validation_test.py` holds every public module-level function whose required parameters are all library input types to the rule of #744: a malformed argument leaves as a BTClibException. One predicate and not a tuple, which is what #743's base class was landed for. It calls with every argument malformed at once, and that is what makes it automatic: no valid values have to be tabulated -- a valid Octets is 20 bytes here, 32 there and any length elsewhere -- and whichever argument the function refuses first, the rule says it must refuse it as a btclib error. Three lists carry what the run finds, and each ratchets one way. `_MALFORMED` is the vocabulary, and a type renamed out of it fails rather than shrinking the walk in silence. `_EXCLUDED` is the nine `is_p2*` predicates, with the reason `script_pub_key._is_funct` already gives. `_OPEN` is what the census of #744 has left, each entry naming the class that escapes; an entry that has become compliant fails the run, as RUF100 fails an unused noqa, so a fix cannot land without deleting its line. What the walk cannot reach is stated rather than omitted: a parameter behind a default is never driven, `hf` and `network` among them, and a function taking a Tx, a Psbt or a callback needs an instance the vocabulary cannot build. It found what the reading missed: `ecc.dleq.verify_proof` answers False for a pub key that is None, where its own comment says a caller error must raise -- the shape #745 closed in five other verifications, in a function written after that census was taken.
This was referenced Aug 13, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The fourth slice of #744.
Stacked on #764, which is
stacked on #763. GitHub
retargets each as its base merges; review the last commit.
What was open
Ten places compared, added to, or read an attribute off an argument
nothing had checked:
TypeErrorabout operands"5" <= 16,"2015" + 1,"hard" <= 0— raised from underneath the library, naming neither the parameter nor the functionAttributeError.tzinfoon a str, andtotal_secondson the int two unix timestamps subtract to — outside both halves of the contract, so nothing a caller is told to catch would have caught itThe guard is
var_int.serialize's, and the vocabulary isutils.is_integer's: a bool is not a number, because it would be theheight one, the block count one, the leaf index one.
The ten
block.bip34_commitment, andBlock.assert_valid_coinbase_heightthrough it
BlockHeader.assert_valid's timestamp — the defaultcheck_validity=Truepath was as exposed as the explicit oneBlockHeader.assert_valid_time'snowmining.mine'smax_triesand its header, whichdataclasses.replaceused to complain about ("should be called ondataclass instances", about a call the caller never made)
proof_of_work.next_bits's two datetimesproof_of_work.retarget_first_heightproof_of_work.hash_rate— a difficulty and atimespan are
float, and an integer is one of those, so the check is"a number, and not a bool"; a block count is a count, so it is
is_integerhashes.merkle_root_from_branch's leaf index, which is whatmerkle_proof.assert_as_validandmerkle_proof.verifyreach itthrough — one fix for three entry points
utils.encode_numWorth a reviewer's eye
BlockHeader._assert_valid_typesis an extraction, not only acheck.
assert_validwas one branch under C901's ten; pyproject'smccabe comment prefers a refactor to a
noqa. It gathers the twois_integerchecks that were already there with the new datetime one.# type: ignore[unreachable], on the timestamp check:self.timeis annotateddatetime, so mypy cannot see the callerthis exists for. Local, at the one real exception, rather than a
blanket disable.
merkle_proof.verify's comment stays true. It says aTypeErroris a caller error and not a verdict;
BTClibTypeErroris aTypeError, so it still propagates through theexcept ValueErrorunchanged — the class is narrower, the control flow identical.
_CASESintests/integer_policy_test.py, beside thetwenty-five already there, with their positive assertions: a test that
only checks refusals passes just as well when the field refuses
everything.
hash_rate(1, 600) == hash_rate(1.0, 600.0)is the onethat pins the int-is-a-number half.
Gates
uv run pytest— 26505 passed, coverage 100.00%uv run pre-commit run --all-files— exit 0sphinx-build -W --keep-going— exit 0Summary by Sourcery
Add explicit type validation across block, proof-of-work, hashing, mining, and utility APIs so incorrect argument types fail with BTClibTypeError instead of leaking bare TypeError/AttributeError from underlying operations.
Enhancements:
Documentation:
Tests: