diff --git a/Makefile b/Makefile
index 5ad5ced35..b1dcf0876 100644
--- a/Makefile
+++ b/Makefile
@@ -253,7 +253,9 @@ src/allmydata/_version.py:
 # After changes:
 # `$ make .tox/make-test-py3-all.diff`
 $(foreach side,old new,.tox/make-test-py3-all-$(side).log):
-	(tox --develop -e py36-coverage allmydata || true) | \
+	tox --develop --notest -e py36-coverage
+	(make VIRTUAL_ENV=.tox/py36-coverage TEST_SUITE=allmydata test-venv-coverage \
+		|| true) | \
 		sed -E 's/\([0-9]+\.[0-9]{3} secs\)/(#.### secs)/' | tee "$(@)"
 .tox/make-test-py3-all.diff: .tox/make-test-py3-all-new.log
 	(diff -u "$(<:%-new.log=%-old.log)" "$(<)" || true) | tee "$(@)"
diff --git a/tox.ini b/tox.ini
index 46d3e4cba..c9ec2bb1d 100644
--- a/tox.ini
+++ b/tox.ini
@@ -64,7 +64,9 @@ commands =
 
          # measuring coverage is somewhat slower than not measuring coverage
          # so only do it on request.
-         coverage: make -e TEST_SUITE={posargs:{env:TEST_SUITE}} test-venv-coverage
+         coverage: coverage run -m twisted.trial {env:TAHOE_LAFS_TRIAL_ARGS:--rterrors --reporter=timing} {posargs:{env:TEST_SUITE}}
+         coverage: coverage combine
+         coverage: coverage xml
 
 
 [testenv:integration]