Re: [Cfrg] BLS standard draft

[David Wong <[e-mail]>]

Hi again,

I am not sure what the process is. If you have a github repo where we
can participate or if you want us to give you feedback here. So here
is some feedback from a first read. Feel free to ignore the bits you
don't agree with of course. Some common themes were:

* Keep in mind that this is for implementations, so remove information
that belongs in a whitepaper
* Make the RFC timeless (we should be able to read it in 5 years and
understand it)
* Set things in stone so that the RFC is actionable, don't make it
vague. If people want to add to it, extensions and updates are
possible later.

And here is the more detailed feedback:

- abstract: re-write with "what is it?" in mind first, history bits
can wait until the introduction. I suggest using developer-friendly
terms like "compression" and define aggregation later if the term is
needed. Example:

    BLS is a digital signature scheme with compression properties.
With a given set of signatures (sig_1, ..., sig_n) anyone can produce
a compressed signature sig_compressed. The same is true for a set of
private keys or public keys, while keeping the connection between sets
(a compressed public key is associated to its compressed public key).
Furthermore, the BLS signature scheme is deterministic, non-malleable,
and efficient. Its simplicity and cryptographic properties allows it
to be useful in a variety of use-cases, specifically when minimal
storage space or bandwidth are required.

- intro:
    - "A signature scheme is a fundamental cryptographic primitive
used on the Internet that is used to protect integrity of
communication" -> not necessarily used on the internet, and not
necessarily for integrity of communications.
    - "2.  Verification requires 2 pairing operations." -> at this
point pairing is not defined, and what does that mean for the
developer? how does it compare to other signature schemes that do not
use pairing?
    - "we believe the scheme will find very interesting applications"
-> too temporal. At some point, it is possible that the scheme will be
popular and this sentence will seem out of place.
    - "the BLS signature scheme is already integrated" -> maybe out of
place (as too temporal as well). If not, sort the list by alphabetical
order, I think no one will mind that.
    - "BLS signatures are used for authenticating transactions as well
as votes during the consensus protocol" -> I suggest we itemize the
different use-cases of BLS (from PKI to blockchain).
- section "1.1.  Terminology"
    - "msg" -> I suggest we change that to "message"
    - "sigma" -> "signature"
    - "signer/verifier/aggregator" do we need roles for these? Can't
we do with just an API ("sign/verify/compress")
    - "P1" is defined but never seem to be used. Am I missing something?
    - I suggest we spell "e()" as "pairing()" in the algorithms, and
define it here
- section "1.2.  Signature Scheme Algorithms and Properties"
    - "A signature scheme comes with" -> "Like most signature schemes,
BLS comes with the following API", this way we can leverage the
reader's knowledge of other signature scheme.
    - "The signing algorithm may be deterministic or randomized,
depending on the scheme" -> as this is a spec, we need to make a
decision here. I think it makes more sense to make it deterministic.
- section "1.2.2.  Security" -> do we need these security properties
in the RFC? It sounds to me like they would belong in a whitepaper
instead.
- section "2."
    - "BLS signatures require pairing-friendly curves" -> I suggest
standardizing BLS with a set of curves. Extensions or updates can
later add more curves if needed.
    - "There are two variants of the scheme" -> It'd be nice if the
two variants were specified in this document, as they both have
use-cases.
    - "Put ... in G1" -> not clear, rephrase
- section "2.1.  Preliminaries". I recommend renaming "suite_string"
to "domain_separator" and having specific values for it instead of
potential values. (We're standardizing something after all, ideally it
should be self-contained)
- section "2.4.  Verify: Signature Verification"
    - "4.  If r*Gamma != 0, output "INVALID" and stop" -> I had heard
a while ago that this membership check was patented for ECDH. Anyone
remembers something like this?
- section "2.5.  Aggregate"
    - it should be "sigma = E1_to_string(string_to_E1(sigma_1) + ... +
string_to_E1(sigma_n))"
    - you specify verifying aggregates of SAME msg and of DIFFERENT
msgs, but only have the aggregate algorithm for SAME msg specified.
- section "2.5.3.  Implementation optimizations". Two things:
    - this should be towards the end of the documentation as these are
optional recommendations. Perhaps after "security recommendations" or
as an appendix
    - is it really wise to have the standard contain this? Available
optimizations may change over time. I've also never seen an RFC
talking about optimizations.
- section "2.7.  Security analysis" -> I don't think this is necessary
to have that in the RFC.
- section "3.1. Verifying public keys"
    - define "G2 membership test"
    - "to prevent rogue key attacks" -> needs a reference
- section "3.4. Randomness considerations" needs a citation, for
example on ECDSA issues when the nonce is repeated
- section "4.  Implementation Status". Standards usually don't refer
to implementations AFAIK. I imagine this is because their state can
change, and new good implementations can arise after the RFC is set in
stone. I think this is good to have in the draft though, so perhaps
add an indication somewhere that this will be deleted in the final
document.
- section "6.  IANA Considerations". Do we need this?
- section "2.6.1.  Preliminaries", "In fact, we will pad each
substring with 0s so that the length of each substring is a multiple
of 8." specify that this is in bits.

Cheers,
David