Gitlab CI Predefined variable ${CI_COMMIT_BEFORE_SHA} sometime returns strange output

Question

I wanted to check for changed files between previous commit occurs and latest commit with

git diff ${CI_COMMIT_BEFORE_SHA} ${CI_COMMIT_SHA} --name-only

For official documentation about

${CI_COMMIT_BEFORE_SHA} : The previous latest commit present on a branch. Is always 0000000000000000000000000000000000000000 in pipelines for merge requests.

So from my assumption, when there are some changes but no merge request occur, that ${CI_COMMIT_BEFORE_SHA} shouldn't be 0000000000000000000000000000000000000000. However, recently I've tested with this following .gitlab-ci.yml

stages:
  - info
  - deploy

variables:
  GIT_STRATEGY: fetch
  GIT_CHECKOUT: "true"

information:
  stage: info
  script:
    - echo ${CI_COMMIT_BRANCH}
    - echo ${CI_COMMIT_BEFORE_SHA}
    - echo ${CI_COMMIT_SHA}
    - export CHANGES=`git diff ${CI_COMMIT_BEFORE_SHA} ${CI_COMMIT_SHA} --name-only`
    - echo ${CHANGES}
  only:
    variables:
      - $CI_COMMIT_BRANCH == "test"
  tags:
    - master

Some scenarios

1. Update .gitlab-ci.yml itself, the ${CI_COMMIT_BEFORE_SHA} was not equal all zero (correct)
2. Create text file and add some contents via Gitlab UI Here's where the issue occurred
   • first edit: After running the pipeline ${CI_COMMIT_BEFORE_SHA} was not equal all zero (correct) and ${CI_COMMIT_SHA} was changed to a new number.
   • second edit: ${CI_COMMIT_BEFORE_SHA} was changed to all zero !? and ${CI_COMMIT_SHA} remain the same as the previous one.

from the second attempt, I just only added words in same text file, No merge occur. Why did the ${CI_COMMIT_BEFORE_SHA} behave like this? I also see this kind of behavior in other project. Isn't it weird? or it's my misunderstanding ?

Here's the output from Gitlab's runner panel

1. First edit

Skipping Git submodules setup
Executing "step_script" stage of the job script
00:01
$ echo ${CI_COMMIT_BRANCH}
test
$ echo ${CI_COMMIT_BEFORE_SHA}
02c9b473c269ff69f54eac42fd5d1c4ace98888e
$ echo ${CI_COMMIT_SHA}
c3d04bd751d4199aa6f2a5c43f3da55100ce42a3
$ export CHANGES=`git diff ${CI_COMMIT_BEFORE_SHA} ${CI_COMMIT_SHA} --name-only`
$ echo ${CHANGES}
test.txt
Cleaning up file based variables
00:00
Job succeeded

2. Second edit

Executing "step_script" stage of the job script
00:00
$ echo ${CI_COMMIT_BRANCH}
test
$ echo ${CI_COMMIT_BEFORE_SHA}
0000000000000000000000000000000000000000
$ echo ${CI_COMMIT_SHA}
62f73dcf2ae8672752fa1ace7a77b97d33ee49df
$ export CHANGES=`git diff ${CI_COMMIT_BEFORE_SHA} ${CI_COMMIT_SHA} --name-only`
fatal: bad object 0000000000000000000000000000000000000000
$ echo ${CHANGES}
Cleaning up file based variables
00:00
Job succeeded

Answer 1

If you look to the documentation page of variables: https://docs.gitlab.com/ee/ci/variables/predefined_variables.html

Then you will see that CI_COMMIT_BEFORE_SHA and CI_COMMIT_SHA are always 0000000000000000000000000000000000000000 in pipelines for merge requests.

That is probably the problem you are facing.

You can create your own variables:

export COMMIT_BEFORE_SHA="$(git rev-parse HEAD~1)"
export COMMIT_SHA="$(git rev-parse HEAD~0)"

So the example will look like:

stages:
  - info
  - deploy

variables:
  GIT_STRATEGY: fetch
  GIT_CHECKOUT: "true"

information:
  stage: info
  script:
    - export COMMIT_BEFORE_SHA="$(git rev-parse HEAD~1)"
    - export COMMIT_SHA="$(git rev-parse HEAD~0)"
    - echo "${COMMIT_BEFORE_SHA}"
    - echo "${COMMIT_SHA}"
    - export CHANGES="`git diff "${COMMIT_BEFORE_SHA}" "${COMMIT_SHA}" --name-only`"
    - echo "${CHANGES}"
  only:
    variables:
      - $CI_COMMIT_BRANCH == "test"
  tags:
    - master

Answer 2

From my own tests, I see CI_COMMIT_BEFORE_SHA be 0000000000000000000000000000000000000000 when I push a new branch (with CI_PIPELINE_SOURCE being push)

As of today (March 2022), the documentation does not list all the cases where CI_COMMIT_BEFORE_SHA== 0000000000000000000000000000000000000000