Multiline strings generate invalid output when whitespace prefixed data

[warmfusion]

Multiline yaml entries with whitespace prefixes generate invalid indentation indicators on dump.

Metadata

• kubesplit version: 0.3.2
• Python version: 3.11.2
• Operating System: Macos / Debian Bullseye

Description

When splitting configuration files that present with existing multiline data blocks with indentation indicators, the generated split versions have the wrong indicators set.

What I Did

Given two configmaps in a yaml file named input.yaml

---
apiVersion: v1
data:
  no_indent_hint: |
    some content here
  no_indent_hint_with_whitespace: |

    some content here
kind: ConfigMap
metadata:
  name: example-no-indicators

---
apiVersion: v1
data:
  indent_2: |2
    some content here

  indent_2_with_whitespace: |2

    some content here
kind: ConfigMap
metadata:
  name: example-with-indicators

When processed through kubesplit ❯ kubesplit -i input.yaml -o example, the generated files for the indicators'

(Shown 'merged' for convenience)

---
apiVersion: v1
data:
  no_indent_hint: |
    some content here
  no_indent_hint_with_whitespace: |4

    some content here
kind: ConfigMap
metadata:
  name: example-no-indicators
---
apiVersion: v1
data:
  indent_2: |
    some content here

  indent_2_with_whitespace: |4

    some content here
kind: ConfigMap
metadata:
  name: example-with-indicators

Observe how the entries with whitespace prefixed in their data blocks are now indicated with |4, even if previously a value of None or 2 was given; this is incorrect, and can be observed with yamllint test;

❯ yamllint example/*
example/11--configmap--example-no-indicators.yml
  8:5       error    syntax error: expected <block end>, but found '<scalar>' (syntax)

example/11--configmap--example-with-indicators.yml
  9:5       error    syntax error: expected <block end>, but found '<scalar>' (syntax)

Additional

When using --no-dash-inwards argument, the entries are valid, which I belive hints that the problem lies in the yamkix function here;

https://github.com/looztra/yamkix/blob/main/src/yamkix/yamkix.py#L62-L69

Specifically, the 'transform' operationl named strip_leading_double_space

[warmfusion]

Interestingly, using ruamel.yaml==0.17.14 results in the |4 stuff going away, but also the required |2 values we have in other places also breaks;

eg;

apiVersion: v1
kind: ConfigMap
metadata:
  name: my-example-with-old-ruamel
data:
  example.conf: |2

    configuration text here 

becomes

...
  example.conf: |

    configuration text here

which is wrong too aparently.

[looztra]

sorry for the horrible latency.

I had to publish a new version of yamkix that now requires ruamel.yaml>=0.17.27
so I will also have a look at this

(thanks for the detailed report)

[looztra]

(to be honest, I'll have to dig into the subject of whitespace prefixed data first, I've never heard of this feature before 😺 )

[looztra]

checked with the updated yamkix version (and ruamel.yaml==0.17.32)

Got

---
apiVersion: v1
data:
  no_indent_hint: |
    some content here
  no_indent_hint_with_whitespace: |2

    some content here
kind: ConfigMap
metadata:
  name: example-no-indicators
---
apiVersion: v1
data:
  indent_2: |
    some content here

  indent_2_with_whitespace: |2

    some content here
kind: ConfigMap
metadata:
  name: example-with-indicators

so indicators have the good value, but always output if one of the document has them.

[looztra]

to be sure to understand the target:

• if there is a indicator, we want to keep it in the output?
• if there is a whitespace but no indicator, we don't want it?

or are we OK if the indicators are explicitely set in the output as long as they are like the ones in the inputs? (that seems fine with latest ruamel.yaml, no?)

ping @warmfusion

[looztra]

A new version of kubesplit (0.3.3) has been published, nothing changed in the code, only the requirements that will make you use an updated ruamel.yaml version