bcode_openai_responses 0.0.1-alpha.0

OpenAI Responses API wire format for bcode
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

//! `OpenAI` billing protocol selection; contains no transport, credentials or catalog behavior.

#[cfg(test)]
mod tests;

use super::usage::{OpenAiUsage, OpenAiUsageContext, normalize_usage};
use bcode_model::{TokenUsage, UsageCaptureSpec, UsageDecoder};
use bcode_session_models::OriginalUsage;

/// Serving differences that affect normalization, explicitly selected by the provider.
#[derive(Debug, Clone, Copy)]
pub enum ResponsesUsageDialect {
    /// Native `OpenAI` API or Codex Responses.
    OpenAi,
    /// Bedrock Mantle Responses with per-bucket cache retention.
    Bedrock,
    /// Chat Completions prompt/completion field aliases.
    ChatCompletions,
}

impl UsageDecoder for ResponsesUsageDialect {
    fn capture_spec(&self) -> UsageCaptureSpec {
        match self {
            Self::OpenAi | Self::Bedrock => UsageCaptureSpec {
                api_shape: "responses",
                containers: &[&["response"], &[]],
                default_source: "usage",
                confirmed_fields: &["model", "service_tier", "prompt_cache_retention"],
                complete_sources: &[
                    "response.completed",
                    "response.done",
                    "response.incomplete",
                    "response.failed",
                ],
            },
            Self::ChatCompletions => UsageCaptureSpec {
                api_shape: "chat_completions",
                containers: &[&[]],
                default_source: "usage",
                confirmed_fields: &["model", "service_tier", "prompt_cache_retention"],
                complete_sources: &["usage"],
            },
        }
    }
    fn observe_json(
        &self,
        _previous: Option<&TokenUsage>,
        usage_json: &str,
        source: &str,
        requested: &std::collections::BTreeMap<String, String>,
        confirmed: &std::collections::BTreeMap<String, String>,
    ) -> Result<TokenUsage, String> {
        self.decode(usage_json, source, requested, confirmed)
    }

    fn normalize(&self, original: &OriginalUsage) -> Result<TokenUsage, String> {
        let report = original.reports.last().ok_or("missing usage report")?;
        self.decode(
            &report.usage_json,
            &report.source,
            &original.requested,
            &report.confirmed,
        )
    }
}

impl ResponsesUsageDialect {
    fn decode(
        self,
        usage_json: &str,
        source: &str,
        requested: &std::collections::BTreeMap<String, String>,
        confirmed: &std::collections::BTreeMap<String, String>,
    ) -> Result<TokenUsage, String> {
        if !self.capture_spec().complete_sources.contains(&source) && source != "usage" {
            return Err("unsupported usage source".into());
        }
        let mut usage: OpenAiUsage =
            serde_json::from_str(usage_json).map_err(|_| "invalid usage report")?;
        if let Some(tier) = confirmed.get("service_tier") {
            usage.service_tier = Some(tier.clone());
        }
        let model = confirmed
            .get("model")
            .or_else(|| requested.get("model"))
            .cloned();
        let retention = confirmed
            .get("prompt_cache_retention")
            .or_else(|| requested.get("prompt_cache_retention"));
        let tier = confirmed
            .get("service_tier")
            .or_else(|| {
                requested.get("service_tier").filter(|tier| {
                    matches!(self, Self::Bedrock) || matches!(tier.as_str(), "default" | "standard")
                })
            })
            .cloned();
        let mut normalized = normalize_usage(
            usage,
            OpenAiUsageContext {
                model,
                service_tier: tier,
                prompt_cache_retention: retention.cloned(),
            },
        );
        if matches!(self, Self::Bedrock) {
            normalized.pricing_context.cache_ttl_seconds =
                retention.and_then(|ttl| match ttl.as_str() {
                    "30m" => Some(1800),
                    "1h" => Some(3600),
                    _ => None,
                });
            if !normalized.has_valid_input_breakdown() {
                normalized.pricing_context.request_input_tokens = None;
            }
            if normalized.details.is_empty() {
                normalized.details = super::usage::text_usage_details(&normalized);
            } else {
                for detail in &mut normalized.details {
                    if matches!(
                        detail.bucket,
                        bcode_model::ModelPricingBucket::CacheReadInput
                            | bcode_model::ModelPricingBucket::CacheWriteInput
                    ) {
                        detail.cache_ttl_seconds = normalized.pricing_context.cache_ttl_seconds;
                    }
                }
            }
        }
        Ok(normalized)
    }
}