fugue / goldplate Goto Github PK

0.2.1 has a release here, dated 2022-04-11: https://github.com/fugue/goldplate/releases/tag/v0.2.1
But it seems it has been forgotten to push it to Hackage.

I'd say current master can be released to Hackage as 0.2.1 and the tag can be moved. The only difference seems to be the bump to aeson: 0.2.1...master

Blockers:

• audreyt/regex-pcre-builtin#21
• simonmar/async#133

I managed to build and test goldplate with the following cabal.project.local. So, it seems goldplate is GHC-9.4 ready, we only need the ecosystem to catch up.

allow-newer: *:template-haskell

allow-newer: hashable:*
allow-newer: regex-pcre-builtin:base
allow-newer: async:base
allow-newer: scientific:base
allow-newer: integer-logarithms:*
allow-newer: aeson:*
allow-newer: indexed-traversable:base
allow-newer: indexed-traversable-instances:base
allow-newer: time-compat:base
allow-newer: these:base
allow-newer: attoparsec:*
allow-newer: assoc:base
allow-newer: text-short:*
allow-newer: semialign:base
allow-newer: data-fix:base
allow-newer: splitmix:base
allow-newer: OneTuple:base

constraints: aeson >= 2.1

@jaspervdj-luminal I'd like goldplate to be in stackage, so I filed commercialhaskell/stackage#6052

There I (implicitly) volunteered as stackage maintainer for goldplate, but I am happy to leave this role to you if you are willing to take it.

It would be nice if

cabal test

ran goldplate's own testsuite. I suppose this would be easy to implement by adding a suitable test-suite stanza to the .cabal file.

The benefit would be that users of goldplate could easily learn how to integrate goldplate into their own .cabal-files, by copy-pasting that stanza.

https://github.com/fugue/goldplate/blob/0e847f753c765e6e6e23fa26aa3b3d8e4b5a21b3/tests/echo.goldplate#L6

It is the first example in the docs, so I dare to ask - what does the variable in the headline contain? Is this example self-fulfilling(any output will be considered valid)?

Assertion 'created_directory' eagerly removes directory before other assertions are checked.
E.g. you cannot place the created_directory assertion before the created_file assertion here:
https://github.com/fugue/goldplate/blob/a2c20310149a69522d0efc2cdcd8e1081f5f93b4/tests/create.goldplate#L9-L13

A bit too imperative for my taste. I would have expected a declarative approach to these assertions, so that the order in which I write these do not matter.

To keep goldplate in Stackage, we need to allow aeson-2. I can make the necessary revisions on hackage, but atm, we are blocked on:

• informatikr/aeson-pretty#36 (informatikr/aeson-pretty#37)

$ goldplate tests
...
not ok tests/stdin.goldplate: stdout
...

https://github.com/fugue/goldplate/blob/91748fa2ac52178cb6c5cee32b8f5a022d5bc759/tests/stdin.stdout#L1

On macOS, echo -n "hi" | wc -c outputs some spaces before the letter count:

$ echo -n "hi" | wc -c 
       2
$ echo -n "hi" | wc -c | hexdump
0000000 20 20 20 20 20 20 20 32 0a                     
0000009

I get not ok ... stdout verdicts under Windows when the same tests pass under Linux and macOS.
I'd guess these are caused by directory globbing resolved under Windows using backslashes instead of slashes, so printing the filepath to stdout gives different results under Linux and Windows. https://github.com/andreasabel/miniagda/blob/67f9ad914f8bbda153d679a517da98b01e482d4e/test/fail/adm/adm1.err#L1
Since Windows nowadays understands the slash as path separator, maybe the slash could be used instead of the backslash to make goldplate more portable?

I accidently put this in my .goldplate file:

{ "command"     : "runghc"
, "input_files" : "*.hs"
, "arguments"   : [ "${GOLDPLATE_INPUT_FILE}" ]
, "asserts"     :
  [ { "exit_code" : 0 
    , "stdout"    : "${GOLDPLATE_INPUT_NAME}.golden" 
    }
  ]
}

Correct would have been:

, "asserts"     :
  [ { "exit_code" : 0 }
  , { "stdout"    : "${GOLDPLATE_INPUT_NAME}.golden" }
  ]

goldplate simply ignored the stdout part. I would have preferred an error message alerting me of my mistake.

Blocked on:

• aeson (haskell/aeson#885)
• semialign (is actively worked on)

I tried

$ cabal build --allow-newer=aeson:all --constraint='aeson==2.*' \
  --allow-newer=scientific:{base,template-haskell} --constraint='scientific==0.3.7.*' \
  --allow-newer=semialign:base 

but semialign does not build with 9.2 because Option was removed from base-4.16.

goldplate builds with cabal build --allow-newer.

• Outdated upper bounds can be relaxed as follows:

     aeson-pretty         >= 0.8  && < 0.9,
     async                >= 2.2  && < 2.3,
     base                 >= 4.9  && < 5,
-    bytestring           >= 0.10 && < 0.11,
+    bytestring           >= 0.10 && < 0.12,
     Diff                 >= 0.3  && < 0.5,
     directory            >= 1.3  && < 1.4,
     filepath             >= 1.4  && < 1.5,
     Glob                 >= 0.10 && < 0.11,
-    optparse-applicative >= 0.14 && < 0.16,
+    optparse-applicative >= 0.14 && < 0.17,
     process              >= 1.6  && < 1.7,
     regex-pcre-builtin   >= 0.95.1.3 && < 0.96,
     text                 >= 1.2  && < 1.3,
-    time                 >= 1.8  && < 1.10,
+    time                 >= 1.8  && < 1.12,
     unordered-containers >= 0.2  && < 0.3

• Regular build is currently blocked by audreyt/regex-pcre-builtin#18.

I'm trying to package goldplate in Debian. I've an issue with regex-pcre-builtin, which is not packaged (and, from what I understood, not wanted) in Debian. regex-pcre is already packaged, but it does not seem to have Text.Regex.PCRE.Text. Is there a reason why Text are used in src/Text/Regex/PCRE/Simple.hs, and not Bytestring? It would help packaging a lot if I could use regex-pcre instead of regex-pcre-builtin...