server/ci: rework version baking, and cache dist-newstyle in CI

UPDATE: After testing in CI it turns out that the compile time Improvement is better than expected: even though we always have to recompile the OSS lib (due to Version.hs), downstream packages like Pro and multi-tenant can still benefit from some caching and avoid full recompilation.  In the best case this takes us from 22 minutes to 13 minutes total.

PR-URL: https://github.com/hasura/graphql-engine-mono/pull/4104
GitOrigin-RevId: 76cbfc157064b33856e30f4c2b2ab2366f9c6089

scripts/dev.sh

@@ -148,6 +148,12 @@ MODE="$1"
 PROJECT_ROOT="$( cd "$( dirname "${BASH_SOURCE[0]}" )/.." >/dev/null 2>&1 && pwd )"   # ... https://stackoverflow.com/a/246128/176841
 cd "$PROJECT_ROOT"
 
+# In CI we use the get version script to actually populate the version number
+# that will be compiled into the server. For local development we use this
+# magic number, which means we won't recompile unnecessarily. This number also
+# gets explicitly ignored in the version test in integration tests. 
+echo '12345' > "$PROJECT_ROOT/server/CURRENT_VERSION"
+
 # Use pyenv if available to set an appropriate python version that will work with pytests etc.
 if command -v pyenv >/dev/null; then
   # For now I guess use the greatest python3 >= 3.5

server/graphql-engine.cabal

@@ -10,6 +10,22 @@ maintainer:          vamshi@hasura.io
 copyright:           Hasura Inc.
 category:            Database
 build-type:          Simple
+extra-source-files:
+  -- We use TH to bake in the server’s version number at compile time. In order
+  -- for recompilation detection to work correctly (especially in the presence
+  -- of caching) we need to both communicate this data via a file (referenced in
+  -- TH with addDependentFile) /and/ add that file to this section of the cabal
+  -- file. See: https://github.com/haskell/cabal/issues/4746 
+  --
+  -- This file is intentionally .gitignore'd
+  CURRENT_VERSION
+  -- These are files referenced by functions from 'file-embed' which uses
+  -- addDependentFile internally and has the same issue as above:
+  src-rsr/mysql_table_metadata.sql
+  src-rsr/mssql_table_metadata.sql
+  src-rsr/catalog_versions.txt
+  src-rsr/catalog_version.txt
+
 
 source-repository head
   type:     git

server/src-lib/Hasura/Server/Utils.hs

@@ -1,4 +1,4 @@
-{-# LANGUAGE TemplateHaskell #-}
+{-# LANGUAGE TemplateHaskellQuotes #-}
 
 module Hasura.Server.Utils
   ( APIVersion (..),
@@ -16,7 +16,6 @@ module Hasura.Server.Utils
     fmapL,
     generateFingerprint,
     getRequestHeader,
-    getValFromEnvOrScript,
     gzipHeader,
     httpExceptToJSON,
     isReqUserId,
@@ -51,12 +50,10 @@ import Data.ByteString.Base16 qualified as Base16
 import Data.ByteString.Lazy qualified as BL
 import Data.CaseInsensitive qualified as CI
 import Data.Char
-import Data.FileEmbed (makeRelativeToProject)
 import Data.HashSet qualified as Set
 import Data.List.NonEmpty qualified as NE
 import Data.Text qualified as T
 import Data.Text.Extended
-import Data.Text.IO qualified as TI
 import Data.Time
 import Data.UUID qualified as UUID
 import Data.UUID.V4 qualified as UUID
@@ -65,13 +62,9 @@ import Database.PG.Query qualified as Q
 import Hasura.Base.Instances ()
 import Hasura.Prelude
 import Language.Haskell.TH.Syntax (Q, TExp)
-import Language.Haskell.TH.Syntax qualified as TH
 import Network.HTTP.Client qualified as HC
 import Network.HTTP.Types qualified as HTTP
 import Network.Wreq qualified as Wreq
-import System.Environment
-import System.Exit
-import System.Process
 import Text.Regex.TDFA qualified as TDFA
 import Text.Regex.TDFA.ReadRegex qualified as TDFA
 import Text.Regex.TDFA.TDFA qualified as TDFA
@@ -127,13 +120,12 @@ parseStringAsBool t
         ++ show falseVals
         ++ ". All values are case insensitive"
 
--- Get an env var during compile time
-getValFromEnvOrScript :: String -> FilePath -> Q (TExp String)
-getValFromEnvOrScript var file = do
-  maybeVal <- TH.runIO $ lookupEnv var
-  case maybeVal of
-    Just val -> [||val||]
-    Nothing -> runScript file
+{- NOTE: Something like this is not safe in the presence of caching. The only
+    way for metaprogramming to depend on some external data and recompile
+    properly is via addDependentFile and to include that file in the
+    extra-source-files in the cabal file (see: https://github.com/haskell/cabal/issues/4746).
+    Leaving this here commented in order to document that fact and also in case
+    there's a way forward in the future.
 
 -- Run a shell script during compile time
 runScript :: FilePath -> Q (TExp String)
@@ -151,6 +143,7 @@ runScript file = do
         ++ " and with error : "
         ++ stdErr
   [||stdOut||]
+-}
 
 -- | Quotes a regex using Template Haskell so syntax errors can be reported at compile-time.
 quoteRegex :: TDFA.CompOption -> TDFA.ExecOption -> String -> Q (TExp TDFA.Regex)

server/src-lib/Hasura/Server/Version.hs

@@ -7,13 +7,16 @@ module Hasura.Server.Version
   )
 where
 
+import Control.Exception
 import Control.Lens ((^.), (^?))
 import Data.Aeson (FromJSON (..), ToJSON (..))
+import Data.FileEmbed (makeRelativeToProject)
 import Data.SemVer qualified as V
 import Data.Text qualified as T
 import Data.Text.Conversions (FromText (..), ToText (..))
 import Hasura.Prelude
-import Hasura.Server.Utils (getValFromEnvOrScript)
+import Language.Haskell.TH
+import Language.Haskell.TH.Syntax
 import Text.Regex.TDFA ((=~~))
 
 data Version
@@ -42,7 +45,21 @@ currentVersion =
   fromText $
     T.dropWhileEnd (== '\n') $
       T.pack $
-        $$(getValFromEnvOrScript "VERSION" "../scripts/get-version.sh")
+        -- NOTE: This must work correctly in the presence of a caching! See
+        -- graphql-engine.cabal (search for “CURRENT_VERSION”) for details
+        -- about our approach here. We could use embedFile but want a nice
+        -- error message
+        $( do
+             versionFileName <- makeRelativeToProject "CURRENT_VERSION"
+             addDependentFile versionFileName
+             let noFileErr =
+                   "\n==========================================================================="
+                     <> "\n>>> DEAR HASURIAN: The way we bake versions into the server has "
+                     <> "\n>>> changed; You'll need to run the following once in your repo to proceed: "
+                     <> "\n>>>  $ echo 12345 > \"$(git rev-parse --show-toplevel)/server/CURRENT_VERSION\""
+                     <> "\n===========================================================================\n"
+             runIO (readFile versionFileName `onException` error noFileErr) >>= stringE
+         )
 
 -- | A version-based string used to form the CDN URL for fetching console assets.
 consoleAssetsVersion :: Text

server/tests-py/test_version.py

@@ -12,6 +12,10 @@ class TestServerVersion(object):
         assert resp.status_code == 200, resp
         my_json = resp.json()
         assert isinstance(my_json, dict), my_json
+
+        # The magic number here means we're compiling for local development and
+        # this test can be ignored:
+        if my_json['version'] != '12345':
         # The tree may be dirty because we're developing tests locally while
         # graphql-engine was built previously when tree was clean. If we're
         # modifying graphql-engine too then both of these will be tagged dirty,