bug: decompress_list raises IndexError on empty string input

[prabindersinghh]

Bug Report

File: metaflow/util.py, line 387

Summary

compress_list([]) legitimately returns "" (joining an empty list).
However, decompress_list("") immediately crashes with IndexError
because lststr[0] is accessed with no guard for empty input.

This makes the natural round-trip decompress_list(compress_list([]))
crash, breaking the contract that these two functions are inverses of
each other.

Reproducer

from metaflow.util import compress_list, decompress_list

print(compress_list([]))            # returns ""
decompress_list(compress_list([]))  # IndexError: string index out of range

Stack trace
File "metaflow/util.py", line 387, in decompress_list
if lststr[0] == zlibmarker:

IndexError: string index out of range

**Expected behavior**

`decompress_list(compress_list([]))` returns `[]`.

**Actual behavior**

`IndexError: string index out of range` at `lststr[0]`.

**Root cause**

Line 387 accesses `lststr[0]` with no guard for empty string.
An empty task ID list is a valid real-world state.

**Environment**
- Metaflow: latest master
- Python: 3.11+

[khushthecoder]

@saikonen I carefully reviewed this issue and understood its details. The key points are as follows:

Why This Issue Matters

The Bug

compress_list([]) legitimately returns "" (an empty list joined with a separator). But decompress_list("") immediately crashes:

>>> from metaflow.util import compress_list, decompress_list
>>> compress_list([])
''
>>> decompress_list('')
# IndexError: string index out of range  (at lststr[0])

This breaks a fundamental invariant: decompress_list(compress_list(x)) == x must hold for all valid inputs, including x = [].

---

Where It Can Surface

compress_list / decompress_list are used to pass input paths — the task IDs that a step depends on — across CLI options and orchestration boundaries:

Call site	Current defense
metaflow/cli_components/step_cmd.py (lines 150, 281)	if input_paths else [] — guards it
metaflow/plugins/argo/conditional_input_paths.py (line 15)	No guard — passes decoded string directly
metaflow/cli.py (import)	Used elsewhere in CLI plumbing

The Argo conditional_input_paths.py call site is particularly concerning — it calls decompress_list(decoded) with no emptiness check. If an Argo workflow ever produces an empty input_paths string (e.g., all paths filtered out by the conditional logic on line 21), this would crash the workflow at runtime.

Even where call sites currently have guards (if input_paths else []), the utility function itself should be robust. External users and internal integrations should not need to know about this landmine.

Semantic Correctness

There's also a subtle correctness issue: if the empty string didn't crash, the current code would reach decoded.split(",") which returns [""] — a list with one empty-string element. But the correct inverse of compress_list([]) is [] (an empty list), not [""]. So the fix needs an explicit early return, not just a crash guard.

---

Proposed Implementation

1. One-line fix in decompress_list (metaflow/util.py)

def decompress_list(lststr, separator=",", rangedelim=":", zlibmarker="!"):
    if lststr == "":          # ← NEW: empty string → empty list (inverse of compress_list([]))
        return []
    # Three input modes:
    if lststr[0] == zlibmarker:
        ...

Why == "" instead of not lststr: Using strict equality avoids silently accepting None, which no current call site passes. If None slips through, it should still fail loudly so the real bug is caught upstream.

---

2. Unit tests (test/unit/test_compress_decompress_list.py)

Test case	Asserts
Empty list compress	compress_list([]) == ""
Empty string decompress (the fix)	decompress_list("") == []
Round-trip empty	decompress_list(compress_list([])) == []
Round-trip simple list	decompress_list(compress_list(["a","b"])) == ["a","b"]
Round-trip prefix-encoded	Verifies : encoding path for common-prefix items
Round-trip zlib-compressed	Forces zlibmin=0 to exercise !-prefixed path
Single element	decompress_list(compress_list(["x"])) == ["x"]

---

3. No changes needed at call sites

This is a backwards-compatible fix. Existing call sites that guard with if input_paths else [] will continue to work. The Argo call site (conditional_input_paths.py) will become safe without needing its own guard.
I'd like to submit a PR for this. Let me know if the approach looks good or if you'd prefer a different style (e.g., not lststr to also accept None).