AesonException when downloading nightly since 2017-08-13

[YoEight]

Greetings,

when stack build I got that output:

Downloaded nightly-2017-08-14 build plan.    
AesonException "Error in $.packages.cassava.constraints.flags['bytestring--lt-0_10_4']: Invalid flag name: \"bytestring--lt-0_10_4\""

I got this since nightly-2017-08-13.

It works if I switch to nightly-2017-08-12.

stack version: 1.5.1

[hvr]

Sounds like Stack reinvents its own parsing...

[snoyberg]

@hvr Perhaps you could help by providing a reference to the Cabal specification that states the rules for flag name parsing? I was unable to find such a document. I'm also certain that, in the spirit of assistance, you'd be happy to change the flag name in cassava in such a way that conforms to the current parser in Stack, as obviously that will have no negative impact on your workflow.

[snoyberg]

@YoEight Herbert is correct, Stack implements its own parsing of flag names (and a few other fields). I've pushed a commit that seems to bring the parsing in line with Cabal's parsing, though it's just a guess based on looking at the code as I have not found any official flag name spec.

Thank you for the report, we'll get this fixed at the Stackage level, I've already pinged the curator on duty right now.

[YoEight]

@snoyberg @hvr Thanks for the quick fix 👍

[snoyberg]

Just to be clear: you still won't be able to use those snapshots unless you use the commit I pushed to Stack (which isn't even on master yet). But I'm sure Herbert will be happy to help with a quick release to Hackage so that you'll be able to get the latest cassava in an upcoming nightly.

[YoEight]

Thanks. This isn't a big issue for me. I was trying to fix my stack nightly resolver build for a lib I maintain. I'm fine with a 2 days old nightly.

[hvr]

@snoyberg You'll surely understand I don't see any point to ineffectively workaround non-compliance bugs in Stack/Stackage, when the reference implementation in Cabal/Hackage works just fine, and would actually have a negative impact on Cabal/Hackage (that's all I'm gonna say here; everything beyond that would require a lengthy blogpost to elaborate). I do agree though, that the specification of valid flag names isn't spelled out in the docs, it's currently basically the regexp [A-Za-z0-9_-]+ with case-insensitive equality. I'll try to remedy that asap. (And fwiw, I'm considering disallowing leading -s)

[snoyberg]

No, I surely do not understand. Can you explain why you are unwilling to make a trivial tweak which will have zero negative impact on cabal users and a positive impact for Stack users?

[chrisdone]

Stack just had a better parser, which handled the separators as actual separators (- and _), naturally disallowing consecutive occurrences. The fact that Cabal's parser didn't bother to validate that properly doesn't really legitimize your use of it here, if anything it shows that Stack flagged up a bug with Cabal's flag parsing. Seeing as you refuse to fix this weirdness, we've lobotomized the parser appropriately.

Package names are also validated,

$ stack new foo--bar
Expected valid package name, but got: foo--bar
Package names consist of one or more alphanumeric words separated by hyphens.
To avoid ambiguity with version numbers, each of these words must contain at least one letter.


Usage: stack new PACKAGE_NAME [--bare] [TEMPLATE_NAME] [-p|--param KEY:VALUE]
                 [DIR] [--solver] [--omit-packages] [--force] [--ignore-subdirs]
                 [--help]
  Create a new project from a template. Run `stack templates' to see available
  templates.

Fortunately this validation is done for package names in Cabal too, as we can see in this user-friendly error message:

$ cabal init --package-name foo--bar
Cannot parse package name: foo--bar
CallStack (from HasCallStack):
error, called at libraries/Cabal/Cabal/Distribution/ReadE.hs:44:24 in
Cabal-1.24.0.0:Distribution.ReadE

So we're safe from @hvr creating package names like cassava--wibble.

[hvr]

@chrisdone Clearly, everyone's entitled to their opinion; however, package name identifiers and flag name identifiers don't use the same syntax, and there's a rationale why package name identifiers are less liberal than flag name identifiers. But your apparent hostility seems to imply that you don't really care to know, so I won't waste both our time to elaborate. (Should I have misread you, and you genuinely want to know, let me know)

[snoyberg]

@hvr I don't see the hostility you're implying @chrisdone showed, he was accurately describing the chain of events here. However, I'm still interested in your response to my question above:

Can you explain why you are unwilling to make a trivial tweak which will have zero negative impact on cabal users and a positive impact for Stack users?