LM tool loop treats max_tool_iterations=1 as immediate failure

[darinkishore]

Summary

LM::execute_tool_loop currently executes the initial tool call, then iterates with for iteration in 1..max_iterations.

When max_tool_iterations is set to 1, that follow-up loop runs zero times and the function always returns Max tool iterations reached, even though one iteration was explicitly allowed.

Why this matters

This makes max_tool_iterations=1 unusable for valid single-iteration tool workflows and can surface as an unexpected hard failure.

Reproduction

1. Configure an LM with max_tool_iterations = 1.
2. Make a request that triggers a tool call.
3. Observe that the initial tool is executed, then the call fails with Max tool iterations reached.

Expected

With max_tool_iterations=1, one follow-up completion turn should be attempted after the initial tool result is appended, or the setting semantics should be clarified and enforced consistently.

Actual

No follow-up completion turn is attempted; the function exits with a max-iterations error.

Candidate fixes

• Adjust loop bounds (for example 1..=max_iterations) if the current setting represents allowed completion iterations after the initial tool call.
• Or redefine semantics so max_tool_iterations is total tool-call budget and update logic/docs accordingly.

[darinkishore]

Confirmed — exact location

crates/dspy-rs/src/core/lm/mod.rs:281:

for iteration in 1..max_iterations {

When max_tool_iterations = 1, max_iterations = 1, so 1..1 is an empty range — zero iterations.

Fix: 1..=max_iterations (inclusive range) if the semantics are "number of follow-up completion turns allowed after the initial tool call".

Alternatively, if max_tool_iterations should mean "total tool-call budget including the initial one", the initial execution already counts as 1, so the loop should be 1..max_iterations (current behavior is correct for that interpretation, but then max_tool_iterations=1 means "execute the tool but don't do a follow-up completion", which is arguably a valid no-op). Clarify the semantics and update the doc comment on the field.