gproxy-protocol 4.0.3

GPROXY v4 connection modelling, operation declarations, vendor wire types and pairwise conversion
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

use crate::{
    transform::{Report, TransformError},
    wire::{claude::generate_content as c, openai::responses as r},
};

/// Real usage facts only for counters absent from the Claude response.
#[derive(Debug, Default, Clone, Copy, PartialEq, Eq)]
pub struct ResponsesUsageFacts {
    pub cache_write_tokens: Option<i64>,
    pub cached_tokens: Option<i64>,
    pub reasoning_tokens: Option<i64>,
}

fn count(value: i64, _field: &str) -> Result<i64, TransformError> {
    Ok(value)
}

fn sum(a: i64, b: i64) -> Result<i64, TransformError> {
    a.checked_add(b)
        .ok_or_else(|| TransformError::invalid_result("usage", "token count overflow"))
}

fn actual(source: Option<i64>, supplied: Option<i64>, field: &str) -> Result<i64, TransformError> {
    count(
        source
            .or(supplied)
            .ok_or_else(|| TransformError::missing_metadata(field))?,
        field,
    )
}

pub(crate) fn to_responses(
    input: c::Usage,
    facts: ResponsesUsageFacts,
    report: &mut Report,
) -> Result<r::ResponseUsage, TransformError> {
    let uncached = count(input.input_tokens, "usage.input_tokens")?;
    let output = count(input.output_tokens, "usage.output_tokens")?;
    let breakdown = input
        .cache_creation
        .flatten()
        .map(|v| {
            sum(
                count(v.ephemeral_1h_input_tokens, "cache_creation.1h")?,
                count(v.ephemeral_5m_input_tokens, "cache_creation.5m")?,
            )
        })
        .map(crate::transform::optional)
        .transpose()?
        .flatten();
    let written = actual(
        input.cache_creation_input_tokens.flatten().or(breakdown),
        facts.cache_write_tokens,
        "usage.cache_creation_input_tokens",
    )?;

    let cached = actual(
        input.cache_read_input_tokens.flatten(),
        facts.cached_tokens,
        "usage.cache_read_input_tokens",
    )?;
    let thinking = actual(
        input
            .output_tokens_details
            .flatten()
            .map(|v| v.thinking_tokens),
        facts.reasoning_tokens,
        "usage.output_tokens_details.thinking_tokens",
    )?;

    let total_input = sum(sum(uncached, written)?, cached)?;
    for (present, name) in [
        (breakdown.is_some(), "cache_creation"),
        (input.fallback_credit.is_some(), "fallback_credit"),
        (input.inference_geo.is_some(), "inference_geo"),
        (input.iterations.is_some(), "iterations"),
        (input.server_tool_use.is_some(), "server_tool_use"),
        (input.speed.is_some(), "speed"),
    ] {
        if present {
            report.omitted(
                format!("usage.{name}"),
                "Responses has no equivalent usage detail",
            );
        }
    }
    Ok(r::ResponseUsage {
        input_tokens: total_input,
        input_tokens_details: r::ResponseInputTokensDetails {
            cache_write_tokens: written,
            cached_tokens: cached,
            rest: Default::default(),
        },
        output_tokens: output,
        output_tokens_details: r::ResponseOutputTokensDetails {
            reasoning_tokens: thinking,
            rest: Default::default(),
        },
        total_tokens: sum(total_input, output)?,
        rest: Default::default(),
    })
}

pub(crate) fn to_claude(input: r::ResponseUsage) -> Result<c::Usage, TransformError> {
    let total_input = count(input.input_tokens, "usage.input_tokens")?;
    let output = count(input.output_tokens, "usage.output_tokens")?;
    let cached = count(
        input.input_tokens_details.cached_tokens,
        "usage.cached_tokens",
    )?;
    let written = count(
        input.input_tokens_details.cache_write_tokens,
        "usage.cache_write_tokens",
    )?;
    let thinking = count(
        input.output_tokens_details.reasoning_tokens,
        "usage.reasoning_tokens",
    )?;

    Ok(c::Usage {
        input_tokens: total_input - cached - written,
        output_tokens: output,
        cache_creation_input_tokens: Some(Some(written)),
        cache_read_input_tokens: Some(Some(cached)),
        output_tokens_details: Some(Some(c::OutputTokensDetails {
            thinking_tokens: thinking,
            rest: Default::default(),
        })),
        cache_creation: None,
        fallback_credit: None,
        inference_geo: None,
        iterations: None,
        server_tool_use: None,
        service_tier: None,
        speed: None,
        rest: Default::default(),
    })
}