diff --git a/.github/workflows/linux.yml b/.github/workflows/linux.yml
index 9ffc609e8..e26d2e9c8 100644
--- a/.github/workflows/linux.yml
+++ b/.github/workflows/linux.yml
@@ -217,8 +217,9 @@ jobs:
       run: |
         export PATH=~/.local/bin:$PATH
         COLUMNS=80 $stack exec -- shelltest --execdir -j16 hledger/test -x /_ -x /addons
-        # XXX run the bin/ func tests corresponding to the GHC version enabled above, only, for now
-        COLUMNS=80 $stack exec -- shelltest --execdir -j16 bin/_scripts8.2.test
+        # XXX run the bin/ func tests corresponding to the GHC version enabled above, only
+        # XXX it's actually too hard to make this reliable for CI right now
+        # COLUMNS=80 $stack exec -- shelltest --execdir -j16 bin/_scripts8.2.test
 
     - name: Test haddock generation
       env:
diff --git a/bin/_scripts8.2.test b/bin/_scripts8.2.test
index 589d3b57c..c4b7d38d9 100644
--- a/bin/_scripts8.2.test
+++ b/bin/_scripts8.2.test
@@ -1,14 +1,5 @@
 # Functional tests for the add-on scripts in this directory.
-# Keep synced: compile.sh, scripts*.test, hledger-*.hs ...
-
-# Check that these ones still compile with this version of hledger.
-# stdout and exit code are ignored for cleaner failure output.
-# 
-# The tests below are hard coded to one GHC version.
-# See also scriptsX.Y.test, used by our CI workflows which run other GHC versions
-# (so they don't install and build for a whole extra GHC version).
-#
-# $ ./compile.sh >/dev/null
+# See scripts.test.
 $  stack --stack-yaml ../stack8.2.yaml ghc hledger-balance-as-budget.hs     >/dev/null || true
 $  stack --stack-yaml ../stack8.2.yaml ghc hledger-check-fancyassertions.hs >/dev/null || true
 $  stack --stack-yaml ../stack8.2.yaml ghc hledger-check-tagfiles.hs        >/dev/null || true
diff --git a/bin/scripts.test b/bin/scripts.test
index 23f545527..b6f3a9eb8 100644
--- a/bin/scripts.test
+++ b/bin/scripts.test
@@ -1,18 +1,18 @@
 # Functional tests for the add-on scripts in this directory.
-# Keep synced: compile.sh, scripts*.test, hledger-*.hs ...
+# Keep synced: compile.sh, scripts*.test, hledger-*.hs..
 
 # Check that these ones still compile with this version of hledger.
 # stdout and exit code are ignored for cleaner failure output.
-# 
-# The tests below are hard coded to one GHC version.
-# See also scriptsX.Y.test, used by our CI workflows which run other GHC versions
-# (so they don't install and build for a whole extra GHC version).
-#
-# $ ./compile.sh >/dev/null
-$  stack ghc hledger-balance-as-budget.hs     >/dev/null || true
-$  stack ghc hledger-check-fancyassertions.hs >/dev/null || true
-$  stack ghc hledger-check-tagfiles.hs        >/dev/null || true
-$  stack ghc hledger-combine-balances.hs      >/dev/null || true
-$  stack ghc --package string-qq hledger-print-location.hs        >/dev/null || true
-$  stack ghc --package string-qq hledger-smooth.hs                >/dev/null || true
-$  stack ghc --package string-qq hledger-swap-dates.hs            >/dev/null || true
+# XXX But installing deps can also generate stderr, causing these to fail.
+# XXX These tests are hard coded to one GHC version.
+# See also _scriptsX.Y.test, used by CI workflows with other GHC versions
+# to avoid installing and building for multiple GHCs.
+# XXX it's actually too hard to make this reliable for CI right now.
+# Maybe still useful for running locally with "make functest".
+$  stack ghc hledger-balance-as-budget.hs                  >/dev/null || true
+$  stack ghc hledger-check-fancyassertions.hs              >/dev/null || true
+$  stack ghc hledger-check-tagfiles.hs                     >/dev/null || true
+$  stack ghc hledger-combine-balances.hs                   >/dev/null || true
+$  stack ghc --package string-qq hledger-print-location.hs >/dev/null || true
+$  stack ghc --package string-qq hledger-smooth.hs         >/dev/null || true
+$  stack ghc --package string-qq hledger-swap-dates.hs     >/dev/null || true