Empty docs break sort_by()

[matthiasvegh]

Describe the bug

If there are empty documents passed into eval-all, sorting only works before and after the empty document.

Version of yq: 4.43.1
Operating system: Ubuntu 20.04
Installed via: wget https://github.com/mikefarah/yq/releases/download/v4.43.1/yq_linux_amd64

Input Yaml
data1.yml:

---
kind: Service
metadata:
  name: b
---
kind: Deployment
metadata:
  name: a
---
# Empty document, meaningful
---
kind: ConfigMap
metadata:
  name: e
---
kind: ConfigMap
metadata:
  name: d

Command
The command you ran:

yq eval-all '[.] | sort_by(.kind, .metadata.name) | .[] | split_doc'

Actual behavior

kind: Deployment
metadata:
  name: a
---
kind: Service
metadata:
  name: b
---

---
kind: ConfigMap
metadata:
  name: d
---
kind: ConfigMap
metadata:
  name: e

Expected behavior

---
kind: ConfigMap
metadata:
  name: d
---
kind: ConfigMap
metadata:
  name: e
---
kind: Deployment
metadata:
  name: a
---
kind: Service
metadata:
  name: b

Additional context
It appears that empty documents behave like an ordering boundary, where everything before it is sorted correctly, as is everything after, but not across. Curiously, if instead of sort_by(.kind, .metadata.name) I do sort_by("\(.kind), \(.metadata.name)") everything is sorted correctly.

My actual motivation is to compare Kubernetes manifests as in #1101. In my case, the manifests are produced by Helm, which can sometimes emit empty documents if the manifest is not needed.

[krokofant]

This became very confusing in our custom diffing workflow when we accidentally left extra --- somewhere in the yaml 😅

Would be nice with a cleaner workaround that more expressively filter out empty documents before doing the sort.

[mikefarah]

Thanks for raising, will fix in the next release

[mikefarah]

Fixed in 4.44.5