Hey Levente,

On 3/20/20 7:19 PM, Levente Uzonyi wrote:

Hi Robert,

On Fri, 20 Mar 2020, Robert wrote:

...

Hey Levente,

On 3/20/20 5:10 PM, Levente Uzonyi wrote:

Hi Robert,

On Fri, 20 Mar 2020, Robert wrote:

Hi Levente,

I got tangled up in an image that had older Crypto code loaded, then I loaded the most recent and I had failing tests. So I went back to a

Interesting. I did my best to minimize such changes, but it's possible that some objects held references to instances of obsolete classes, which is not easy to fix.

Perhaps, I 86ed it and started over.

virgin image and loaded into that. All tests pass Green. I am trying to track all the changes you made, for instance relying on the stock ThirtyTwoBitRegister>>#rotateLeftBy: and also using the native

ThirtyTwoBitRegister was moved to Squeak 10+ years ago. Overriding a method of it from an external package is not a good idea, as the internals of the class may change over time.

I totally agree. I made that mistake with DateAndTime.

If you compare the two versions of #rotateLeftBy:, you'll find that Squeak's version is faster.

Indeed.

Now that we have a dependency on the Registers package, it is possible to migrate all users of ThirtyTwoBitRegister to those registers in the Registers package for an additional performance boost.

Exactly what I was thinking!

SecureHashAlgorithm>>#hashInteger:seed:. It is confusing. Cryptography

It is confusing. The idea of using integers with any HashFunction is simply wrong. HashFunctions are defined on ByteArray inputs, so using integers means some kind of mapping between the two - aka hardcoded endianness. Besides being confusing, this is a great source of performance loss.

Then we should rewrite all senders of such to use the byteArray interface.

I did not want to rewrite the senders of #hashInteger:seed:,

Oh, I think we should!

and I did not intend to introduce the method because of the above, so I decided to do the simplest thing and use SecureHashAlgorithm which is present in Squeak core, provided by the same package ThirtyTwoBitRegister lives in.

I would rather not depend on that, even though in a system category; I'd like self-contained.

I wanted to finish the integration with minimal changes. Even the ThirtyTwoBitRegister stuff was too much as it caused unexpected problems. So, I agree that it's better to not depend on SecureHashAlgorithm, but it takes time and effort to make such changes, so I took the easy path.

It's good, you kept it running, running right, running fast. It is no different than before, depending on SecureHashAlgorithm.

...

was supposed to be complete and independent, but it relies on certain classes being present, with certain implementations. Given this, would

Well, define independend. The package relies on many classes not in the Cryptography repository but the Squeak core (e.g. Integer, ByteArray, ByteString, Stream, ThirtyTwoBitRegister, etc). The ThirtyTwoBitRegister and SecureHashAlgorithm I would rather not be dependent on.

ProCrypto be loadable in Pharo and pass green?

I do not use Pharo, but based on the way Pharo is developed, I doubt it would load and work in any existing Pharo versions except for the most ancient ones.

It used to load.

I think that even if you made it work in the latest Pharo version, it would very likely stop working in the next release. So, if I were you, I wouldn't bother.

I won't bother. There is poison there. Good to check your code, though.

I wouldn't be surprised if they dropped Monticello support in the next release, forcing you to use whatever packaging system they come up with.

One Pharo enthusiast called Monticello config maps "and mcm config files is a thing of the stone age :wink: ". I think not, I really like configuration maps. It gets the job done and from multiple repositories as well!

One thing I recall you saying, I think you said you are using postscript initialization. Where could I find that?

Open the Monticello Browser, select a package, click on Scripts. A menu will appear where you can choose which package script you would like to edit. I added a postscript to CryptographyCiphers to reinitialize the class variables of Rijndael.

Alright, this is the sort of code I like to see in a #startUp: method so I moved it and added Rijndael to the startUpList in #initialize. There is a CryptographyCiphers-rww.19 and the ProCrypto config is up to date with it.

That's not the right place for such code. The package postscript is executed when you load a version of the package that has a different postscript than the previous version you had loaded into your image. Things on the startup list are evaluated every single time you start your image. There's no need to run this code that often. That script simply fixes your image in the extremely rare case when you loaded CryptographyCore-ul.7, then CryptographyCiphers-rww.16, and then CryptographyCiphers-ul.17 in that order.

I understand your point, I went for overkill where every time it starts up it reinitializes. I'm not a big fan of postscript, it is like the initialization code is hiding. I touched it up. I can go either way.

K, r

Levente

...

K, r

Levente

k, r

On 3/18/20 7:50 AM, Levente Uzonyi wrote:

Hi Robert,

I've finished the integration of Hasher into the Cryptography package. I didn't publish Monticello Configurations, but here's a list I would publish if I did:

Registers-Core-ul.1 CryptographyCore-ul.7 CryptographyHashing-ul.23 CryptographyASN1-ul.6 CryptographyRandom-ul.13 CryptographyCiphers-rww.16 CryptographySignatures-ul.17 CryptographyKeyExchange-rww.14 CryptographyArchive-ul.18 CryptographyX509-ul.15

The Registers repository should be added to the configuration first, then you can select Registers-Core and move it to the top of the list. Regarding load order, I swapped CryptographyHashing with CryptographyASN1 because I added extension methods to the latter which extend classes of the former.

Since CryptographyHashing does not depend on CryptographyCore, those two can be swapped as well if needed.

For the tests, the configuration should contain the following packages:

CryptographyASN1Tests-rww.1 CryptographyCoreTests-rww.1 CryptographyHashingTests-ul.2 CryptographyRandomTests-rww.1 CryptographyCiphersTests-rww.1 CryptographySignaturesTests-rww.1 CryptographyKeyExchangeTests-rww.1 CryptographyArchiveTests-rww.1 CryptographyX509Tests-rww.1 Registers-Tests-ul.1

Levente

On Tue, 17 Mar 2020, Robert wrote:

Hi Levente,

On 3/16/20 2:53 PM, Levente Uzonyi wrote:

Hi Robert,

On Fri, 13 Mar 2020, Robert wrote:

On 3/12/20 7:49 PM, Levente Uzonyi wrote:

On Thu, 12 Mar 2020, Robert wrote:

Oh? I thought we discussed your package becoming the core solution, for HashFunctions. I was thinking you were going to rename all your classes back to no prefixes (except RGThirtyTwoBitRegister renamed to CryptoThirtyTwoBitR egister). And your hashFunction becomes the one in CryptographyCore. Your Registers and HashFunctions, become CryptographyHashing. Then I'll reset the dependencies to load yours.

I got a little excited and released Pro Crypto v1.1.1, so with your code it would be ProCrypto v1.2.1.

Did I misunderstand?

I proposed that about three times in these discussions but got no response from you. Since you started integrating SHA512 manually, my impression was that you want to keep the exising classes. If you want Hasher to be merged into Cryptography, then I can give it a try on the weekend, but

See this post: http://lists.squeakfoundation.org/pipermail/squeak-dev/2020-March/207872.htm...

I must have missed that post.

I apologize for tacking after some other stuff, in the middle of a paragraph. That was limiting.

1. HashFunction should stay in the same package as its subclasses. Why?

When another package doesn't use any of HashFunction's subclasses, it will not use HashFunction. When another package needs a subclass, HashFunction has to be present.

No, I see this through the lens of reuse. there may be other packages that wish to extend hashFunction for their own use. AN example would be a SignalEncryption package. It has a HashFunction. So all the class roots belong in the Core package.

I couldn't find the SignalEncryption package anywhere.

It is CryptoChallengeNumberFive. I had downloaded the java implementation of Signal and they have 2 special crypto algorithms, I think one a hashFunction and the other their Public/PrivateKeys. Thus I wrote for NumberFive Sig nalEncryption and SignalProtocol. The first would extend root classes in Core for the HashFunction and the Keys from KeyExchange. Oh! Yeah so we already use a subpackage to subclass Keys, I think. So why not Hashing? Okay, le ave it in CryptographyHashing, I see now.

Is there a cryptographic hash function that is missing from Hasher?

There could be as alternate packages extend it, such as our future SignalEncryption package.

If there is, can it be implemented as a subclass of HashFunction?

Sure, just load Hashing.

If yes, why doesn't it belong to the same package as the other subclasses of HashFunction?

To promote broad development to extend as needed. Perhaps rolled over to Hashing later...Does that work?

2. Registers should stay in a separate package. Why?

They can be used for other things. For example, I've got an unpublished package containing various PRNG implementations using it.

That's fine then, please put these classes also in the Core package.

I don't see how the CryptographyCore package would be a good place for Registers.

Okay, same deal as above then. But you are adding another package dependency. Easy enough to record in the ProCrypto-1-1-1 configuration map.

The PRNGs I implemented are unrelated to Cryptography.

PRNGs? I had not heard. Do you want to roll those over to Random, please?

K, r

Levente

Does that work for you?

Working our way to the garden.

k, r

Levente

k, r

On 3/12/20 7:26 PM, Levente Uzonyi wrote:

Hi Robert,

On Wed, 11 Mar 2020, Robert wrote:

Dear Levente,

I had to rework the Hashing package. It was recording change records that moved RGSixtyTwoBitRegisters before another to rename them CryptoSixtyTwoBitRegisters, CryptographyHashing was ripping them out of your Registers packa ge and your code started failing. So I had to swap classes around packages and fix a few issues I had with SHA512 initialization, class & instance sides. I verified that they load in either order now and fully CryptoGreen. I setup dependencies through the latest Hashing package, 21. Here are the versions & how I load:

Anything with your merge I can help with, Levente? I am excited for the day to announce ProCrypto v1.1.1, you know! ^,^ Milk it. I added a pointer to the plugins.

What merge do you mean?

Levente

Installer ss project: 'Registers'; install: 'Registers'; project: 'Hasher'; install: 'HAHasher-Core'; install: 'HAHasher-Tests'. Installer ss project: 'Cryptography'; install: 'CryptographyP lugins'; install: 'CryptographyX509'.

K, r

ProCrypto packages and dependencies Package Size (kb) Dependencies Algorithms 1 CryptographyCore-rww.5 18 HMAC, CBC, CFB, CTR, OFB 2 CryptographyASN1-rww.4 58 ASN1Module, ASN1InputStream, ASN1OutputStream 3 CryptographyHashing-rww.21 208 CryptographyCore-rww.5 ND2, MD4, MD5, SHA1, SHA256, SHA512 4 CryptographyRandom-rww.11 21 CryptographyHashing-rww.21 RandomPool, PrimesFinder, Miller-Rabin, Fortuna, SecureRandom 5 CryptographyCiphers-rww.15 81 CryptographyRandom-rww.11 CryptographyASN1-rww.4 ARC2, ARC4, DES, TripleDES, Blowfish, Rijndael 6 CryptographySignatures-rww.15 37 CryptographyCiphers-rww.15 DSAKeyPairGenerator, ElGamalKeyPairGenerator, RSAKeyPairGenerator 7 CryptographyKeyExchange-rww.13 5 CryptographySignatures-rww.15 Diffie-Hellman 8 CryptographyArchive-rww.15 17 CryptographyKeyExchange-rww.13 PBKDF2WithHmacSHA1, PBKDF2WithHmacSHA256, PKCS12 9 CryptographyX509-rww.13 34 CryptographyArchive-rww.15 X509Certificate, X509CertificateDerReader, DSAPrivateKeyFileReader, RSAPublicKeyFileGenerator, RSAPrivateKeyFileGenerator 479 Loadable Unloadable

On 3/10/20 8:31 PM, Robert wrote:

     I should share with you that I can load Levente's work in parallel and there are no toes stepped on. And all of his tests are CryptoGreen, with & out. This is a good.

     *message too large* kindly, rabbit

     On 3/10/20 6:06 PM, Robert wrote:

     Hi Levente,

     Here is a new release of CryptographyHashing-rww.15. It is not linked up through dependencies, so load it after. It supports SHA512WithPrimitive and SHA512NonPrimitive and passes all tests. CryptoGreen for SHA512, wit

h the shiny, new SHA2Plugin and without. Find plugins here, for linux64x64: https://www.dropbox.com/home/Callisto%20House/squeak-crypto-plugins .

     Here is this working implementation of SHA512. The naming ought to be without prefix for th ecore classes. I have no problem whatsoever if we were to rebase your work as the defining implementation for all of thosew f

uncrtions, using one plugin. That's something wonderful. We should use you hashFunction and rename without prefix. Tests separate, that's fashionable. We can figure out the mc config later.

     publish your work on, then I will link your solution into dependencies.

                                                                                                                                               CryptographyHashing-ul.16

                                                                                                                                          CryptographyHashing-rww.15 (Release)

     File:
     CryptographyHashing-rww.15.mcz
     Author:
     Robert Withers
     Timestamp:
     10 March 2020 9:57:39 pm
     UUID:
     b7df722e-ab05-4465-97ef-deeffb0212d0
     Ancestors:
     CryptographyHashing-rww.14
     Dependencies:
     CryptographyCore-rww.5
     Release:
     This is a release that can be read by anybody.
     Message:
     adapt to new #primCopyoubleWords:intoBytes:. CryptoGreen for SHA512, with the shiny, new SHA2Plugin and without. Find plugins here, for linux64x64:

https://www.dropbox.com/home/Callisto%20House/squeak-crypto-plugins . rttyk, r

--

--

-- Kindly, Robert

-- Kindly, Robert

-- Kindly, Robert