[ANDROIDHASSAN]

Problem: safe_render's docstring says missing keys become empty strings, and Instructions.render(data=...) repeats that. Three kinds of placeholder raise instead:

Instructions("Hello {name}, total {}").render(data={"name": "Alex"})
# IndexError: tuple index out of range
Instructions("Hello {name}, item {0}").render(data={"name": "Alex"})
# IndexError: tuple index out of range
Instructions("Hello {name}, price {price:.2f}").render(data={"name": "Alex"})
# ValueError: Unknown format code 'f' for object of type 'str'

{} and {0} look up positional args, which are always empty here, so get_field raises IndexError; it only caught KeyError, AttributeError and TypeError. For {price:.2f} the lookup is caught, but the "" stand-in is then formatted with .2f.

Fix: catch IndexError too, and return None for a missing field instead of "". format_field already renders None as empty before applying the spec. convert_field passes None through, so {x!s} stays empty instead of becoming "None".

Tests

test_render_missing_placeholders_become_empty in tests/test_chat_ctx.py, through the public Instructions.render. On main:

FAILED ...[Hello {name}, total {}-Hello Alex, total ]        IndexError: tuple index out of range
FAILED ...[Hello {name}, item {0}-Hello Alex, item ]          IndexError: tuple index out of range
FAILED ...[Hello {name}, price {price:.2f}-Hello Alex, price ] ValueError: Unknown format code 'f' for object of type 'str'
3 failed, 1 passed

The fourth case ({caller!s}) passes on main and is there so returning None can't leak "None" into a prompt.

ruff format, ruff check, make type-check pass. uv run pytest --unit --audio_eot -q: 5045 passed, 5 skipped.

Scope: safe_render only. Placeholders that resolve to a value render as before. One side effect: an explicit None value with !s/!r now renders empty rather than "None", matching the docstring's "None values also become empty strings".

[devin-ai-integration]

🔍 Devin Review: 1 flag

Not posted on this PR by your GitHub settings — view it in Devin Review. (Configure)

[xianshijing-lk]

I think the PR addressed the problems it found, lgtm in general.

Adding @longcw or @theomonnom for a second look to double check if it already covers all the use cases that we need.