harn-vm 0.10.160

Async bytecode virtual machine for the Harn programming language
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184

use crate::call_budget::{charge_mcp_call, install_mcp_call_budget, mcp_calls_spent};
use crate::llm::cost::*;
use crate::orchestration::ambient_scope::{scope_ambient, AmbientExecutionScope};
use crate::value::{ErrorCategory, VmError};
use std::future::Future;
use std::task::{Context, Poll};

fn completed_call() -> crate::llm::api::LlmResult {
    serde_json::from_value(serde_json::json!({
        "text": "ok",
        "tool_calls": [],
        "input_tokens": 3,
        "output_tokens": 2,
        "cache_read_tokens": 1,
        "cache_write_tokens": 0,
        "cache_supported": true,
        "model": "scope-test-unpriced-model",
        "provider": "mock",
        "blocks": [],
        "logprobs": [],
    }))
    .unwrap()
}

fn assert_uncapped() {
    assert_eq!(peek_llm_cost_budget(), None);
    assert_eq!(peek_llm_token_budget(), None);
    assert_eq!(peek_total_cost(), 0.0);
    assert_eq!(peek_total_tokens(), 0);
    assert_eq!(
        peek_observed_session_usage(),
        ObservedSessionUsage::default()
    );
    assert_eq!(mcp_calls_spent(), None);
}

#[tokio::test(flavor = "current_thread")]
async fn interleaved_spans_keep_ceilings_and_accounting_until_their_own_completion() {
    reset_cost_state();
    async fn turn(max: f64, tokens: u64, spent: f64, calls: u64) {
        let _cost = install_llm_cost_budget_seeded(Some(max), spent);
        let _tokens = install_llm_token_budget(tokens);
        let _mcp = install_mcp_call_budget(calls);
        for _ in 0..calls {
            charge_mcp_call().unwrap();
        }
        accumulate_llm_usage("test", 0, 0, 0.125).unwrap();
        for _ in 0..calls {
            record_llm_usage(&completed_call()).unwrap();
        }
        tokio::task::yield_now().await;
        // MCP is the direction control: its existing captured owner must
        // survive the exact same poll order as the newly captured LLM owners.
        assert_eq!(mcp_calls_spent(), Some(calls));
        assert!(charge_mcp_call().is_err(), "each MCP ceiling must fire");
        assert_eq!(peek_llm_cost_budget(), Some(max));
        assert_eq!(peek_llm_token_budget(), Some(tokens));
        assert_eq!(peek_total_cost(), spent + 0.125);
        assert_eq!(peek_total_tokens(), 5 * calls);
        assert_eq!(peek_observed_session_usage().calls, calls);
        assert_eq!(
            peek_observed_session_usage().cache_read_tokens,
            calls as i64
        );
    }

    let mut first = Box::pin(scope_ambient(
        AmbientExecutionScope::default(),
        turn(1.0, 10, 0.25, 1),
    ));
    let mut second = Box::pin(scope_ambient(
        AmbientExecutionScope::default(),
        turn(2.0, 20, 0.5, 2),
    ));
    let waker = futures::task::noop_waker();
    let mut cx = Context::from_waker(&waker);
    assert!(first.as_mut().poll(&mut cx).is_pending());
    assert_eq!(mcp_calls_spent(), None);
    assert!(second.as_mut().poll(&mut cx).is_pending());
    assert_eq!(mcp_calls_spent(), None);
    // Resume the first while the second guard is still held across its await.
    // Without LLM capture, MCP still reads its own ledger but cost reads 2.0.
    assert_eq!(first.as_mut().poll(&mut cx), Poll::Ready(()));
    assert_uncapped();
    assert_eq!(second.as_mut().poll(&mut cx), Poll::Ready(()));
    assert_uncapped();
}

#[tokio::test(flavor = "current_thread")]
async fn fanout_shares_spend_observations_and_rearmed_ceilings() {
    reset_cost_state();
    let _cost = install_llm_cost_budget_seeded(Some(1.0), 0.25);
    let _tokens = install_llm_token_budget(10);
    // Both child kinds belong to the same accounting tree, even on another
    // executor thread. A fresh install inside either child remains independent.
    let inline = AmbientExecutionScope::capture_for_inline_subtask();
    let worker = AmbientExecutionScope::capture_inherited();
    crate::runtime_stack::on_vm_stack(move || {
        tokio::runtime::Builder::new_current_thread()
            .build()
            .unwrap()
            .block_on(scope_ambient(inline, async {
                accumulate_llm_usage("test", 0, 0, 0.125).unwrap();
                record_llm_usage(&completed_call()).unwrap();
                set_llm_cost_budget(Some(0.375));
                set_llm_token_budget(Some(5));
            }));
    });
    scope_ambient(worker, async {
        assert_eq!(peek_total_cost(), 0.375);
        assert_eq!(peek_total_tokens(), 5);
        assert_eq!(peek_observed_session_usage().calls, 1);
        assert_eq!(peek_llm_cost_budget(), Some(0.375));
        assert_eq!(peek_llm_token_budget(), Some(5));
        // Charge beyond both limits. The completed response must enter both
        // ledgers even though token exhaustion takes precedence in the error.
        assert!(matches!(
            accumulate_llm_usage("test", 1, 0, 0.125),
            Err(VmError::CategorizedError {
                category: ErrorCategory::BudgetExceeded,
                ..
            })
        ));
    })
    .await;
    assert_eq!(peek_total_cost(), 0.5);
    assert_eq!(peek_total_tokens(), 6);
    set_llm_token_budget(None);
    assert!(matches!(
        accumulate_llm_usage("test", 0, 0, 0.125),
        Err(VmError::CategorizedError {
            category: ErrorCategory::BudgetExceeded,
            ..
        })
    ));
    assert_eq!(peek_total_cost(), 0.625);
}

#[tokio::test(flavor = "current_thread")]
async fn cancelling_a_suspended_span_leaves_the_callers_budget_owner_intact() {
    reset_cost_state();
    let _cost = install_llm_cost_budget_seeded(Some(3.0), 0.5);
    let _tokens = install_llm_token_budget(30);
    accumulate_llm_usage("test", 5, 0, 0.0).unwrap();
    let mut suspended = Box::pin(scope_ambient(AmbientExecutionScope::default(), async {
        let _cost = install_llm_cost_budget(1.0);
        let _tokens = install_llm_token_budget(10);
        std::future::pending::<()>().await;
    }));
    let waker = futures::task::noop_waker();
    assert!(suspended
        .as_mut()
        .poll(&mut Context::from_waker(&waker))
        .is_pending());
    drop(suspended);
    assert_eq!(peek_llm_cost_budget(), Some(3.0));
    assert_eq!(peek_llm_token_budget(), Some(30));
    assert_eq!(peek_total_cost(), 0.5);
    assert_eq!(peek_total_tokens(), 5);
}

#[tokio::test(flavor = "current_thread")]
async fn uncapped_fanout_still_has_one_accounting_owner() {
    reset_cost_state();
    let inline = AmbientExecutionScope::capture_for_inline_subtask();
    let worker = AmbientExecutionScope::capture_inherited();
    scope_ambient(inline, async {
        record_llm_usage(&completed_call()).unwrap();
        accumulate_llm_usage("test", 0, 0, 0.125).unwrap();
    })
    .await;
    scope_ambient(worker, async {
        assert_eq!(peek_total_tokens(), 5);
        assert_eq!(peek_total_cost(), 0.125);
        record_llm_usage(&completed_call()).unwrap();
    })
    .await;
    assert_eq!(peek_total_tokens(), 10);
    assert_eq!(peek_total_cost(), 0.125);
    assert_eq!(peek_observed_session_usage().calls, 2);
    assert_eq!(peek_llm_cost_budget(), None);
    assert_eq!(peek_llm_token_budget(), None);
    reset_cost_state();
}