makefile improvement

[scott-cotton]

The introduction of the use of git diff-index --quiet HEAD in the make target tag-release was incorrectly causing pushes of tags with the wrong image tag. I think this was the root cause of https://github.com/signadot/signadot/issues/4587 (although the image tags present in kustomize sources at e2e-v2.X are fixed by https://github.com/signadot/signadot/issues/4585, this PR fixes the cause)

This PR makes tag-release idempotent, rather than reverting the call to git diff-index, as it seems was intended by the introduction of git diff-index .

One other gotcha I found was that git diff-index --exit-code can indicate there are differences when those differences (eg file modification times) don't necessarily cause git commit to commit changes, so git diff --exit-code is used in the replacement script here.

[daniel-de-vera]

The introduction of the use of git diff-index --quiet HEAD in the make target tag-release was incorrectly causing pushes of tags with the wrong image tag

@scott-cotton, I'm not sure I understand this, can you give explain it more deeply? (or give an example?)

[scott-cotton]

The introduction of the use of git diff-index --quiet HEAD in the make target tag-release was incorrectly causing pushes of tags with the wrong image tag

@scott-cotton, I'm not sure I understand this, can you give explain it more deeply? (or give an example?)

According to the docs, git diff-index --quiet implies git diff-index --exit-code, which means that when there is a detected difference, it exits with code 1. The command in the Makefile

git diff-index --quiet HEAD || git commit -m tag-release-$(RELEASE_TAG) k8s/base

Thus never commits the kustomize changes from the previous line and yet it pushes the tag as if it had.

[daniel-de-vera]

According to the docs, git diff-index --quiet implies git diff-index --exit-code, which means that when there is a detected difference, it exits with code 1. The command in the Makefile

git diff-index --quiet HEAD || git commit -m tag-release-$(RELEASE_TAG) k8s/base

Thus never commits the kustomize changes from the previous line and yet it pushes the tag as if it had.

Mmm, check this example:

ddv@ddvpc:/tmp$ git clone [email protected]:signadot/hotrod.git
Cloning into 'hotrod'...
remote: Enumerating objects: 1970, done.
remote: Counting objects: 100% (1315/1315), done.
remote: Compressing objects: 100% (677/677), done.
remote: Total 1970 (delta 694), reused 1052 (delta 557), pack-reused 655
Receiving objects: 100% (1970/1970), 16.30 MiB | 4.39 MiB/s, done.
Resolving deltas: 100% (895/895), done.
ddv@ddvpc:/tmp$ 
ddv@ddvpc:/tmp$ 
ddv@ddvpc:/tmp$ cd hotrod/
ddv@ddvpc:/tmp/hotrod$ 
ddv@ddvpc:/tmp/hotrod$ 
ddv@ddvpc:/tmp/hotrod$ 
ddv@ddvpc:/tmp/hotrod$ export RELEASE_TAG=v1.0.0
ddv@ddvpc:/tmp/hotrod$ 
ddv@ddvpc:/tmp/hotrod$ (cd k8s/base && kustomize edit set image signadot/hotrod:$RELEASE_TAG)
ddv@ddvpc:/tmp/hotrod$ 
ddv@ddvpc:/tmp/hotrod$ git diff
diff --git a/k8s/base/kustomization.yaml b/k8s/base/kustomization.yaml
index f65673c..1c0c3d1 100644
--- a/k8s/base/kustomization.yaml
+++ b/k8s/base/kustomization.yaml
@@ -11,4 +11,4 @@ resources:
 
 images:
 - name: signadot/hotrod
-  newTag: latest
+  newTag: v1.0.0
ddv@ddvpc:/tmp/hotrod$ 
ddv@ddvpc:/tmp/hotrod$ 
ddv@ddvpc:/tmp/hotrod$ git diff-index --quiet HEAD || git commit -m tag-release-$RELEASE_TAG k8s/base
[main 9b4363b] tag-release-v1.0.0
 1 file changed, 1 insertion(+), 1 deletion(-)
ddv@ddvpc:/tmp/hotrod$ 
ddv@ddvpc:/tmp/hotrod$ 
ddv@ddvpc:/tmp/hotrod$ git log
commit 9b4363b652d90542498f22a2e83b1ddc3fcda592 (HEAD -> main)
Author: Daniel De Vera <[email protected]>
Date:   Thu Jun 20 12:10:24 2024 -0300

    tag-release-v1.0.0

commit 16ea7c27c03ac4d4cdb9df72e7996cf446cceba1 (origin/main, origin/HEAD)
Merge: 6142a19 74613ff
Author: Daniel De Vera <[email protected]>
Date:   Wed Jun 19 16:06:38 2024 -0300

    Merge pull request #240 from signadot/update-cypress-test
    
    Update of cypress tests + makefile fix

[scott-cotton]

Well, I must have been mistaken. I had some runs which led me to believe the above analysis (I did a similar thing to your test with different results) but now I am unable to repro.

Apologies for the associated noise.

However, leaving PR open for

• improved idempotency
• bug fix using git diff instead of git diff-index

[daniel-de-vera]

LGTM

[daniel-de-vera]

@scott-cotton, can we merge this PR?