feat(nws, normalizers): add NWS hourly forecast normalization and enforce canonical float rounding

- Implement full NWS hourly forecast normalizer (raw.nws.hourly.forecast.v1 → weather.forecast.v1)
- Add GeoJSON forecast types and helpers for NWS gridpoint hourly payloads
- Normalize temperatures, winds, humidity, PoP, and infer WMO condition codes from forecast text/icons
- Treat forecast IssuedAt as EffectiveAt for stable, dedupe-friendly event IDs

- Introduce project-wide float rounding at normalization finalization
  - Round all float values in canonical payloads to 2 decimal places
  - Apply consistently across pointers, slices, maps, and nested structs
  - Preserve opaque structs (e.g., time.Time) unchanged

- Add SchemaRawNWSHourlyForecastV1 and align schema matching/comments
- Clean up NWS helper organization and comments
- Update documentation to reflect numeric wire-format and normalization policies

This establishes a complete, deterministic hourly forecast pipeline for NWS
and improves JSON output stability across all canonical weather schemas.
This commit is contained in:
2026-01-16 10:28:32 -06:00
parent 0fcc536885
commit 2eb2d4b90f
11 changed files with 831 additions and 92 deletions

View File

@@ -1,51 +1,129 @@
// FILE: internal/sources/nws/forecast.go
package nws
import (
"context"
"fmt"
"encoding/json"
"strings"
"time"
"gitea.maximumdirect.net/ejr/feedkit/config"
"gitea.maximumdirect.net/ejr/feedkit/event"
nwscommon "gitea.maximumdirect.net/ejr/weatherfeeder/internal/providers/nws"
"gitea.maximumdirect.net/ejr/weatherfeeder/internal/sources/common"
"gitea.maximumdirect.net/ejr/weatherfeeder/internal/standards"
)
// ForecastSource polls an NWS forecast endpoint (narrative or hourly) and emits a RAW forecast Event.
//
// It intentionally emits the *entire* upstream payload as json.RawMessage and only decodes
// minimal metadata for Event.EffectiveAt and Event.ID.
//
// Output schema (current implementation):
// - standards.SchemaRawNWSHourlyForecastV1
type ForecastSource struct {
name string
url string
userAgent string
http *common.HTTPSource
}
func NewForecastSource(cfg config.SourceConfig) (*ForecastSource, error) {
if strings.TrimSpace(cfg.Name) == "" {
return nil, fmt.Errorf("nws_forecast: name is required")
}
if cfg.Params == nil {
return nil, fmt.Errorf("nws_forecast %q: params are required (need params.url and params.user_agent)", cfg.Name)
const driver = "nws_forecast"
// NWS forecast endpoints are GeoJSON (and sometimes also advertise json-ld/json).
hs, err := common.NewHTTPSource(driver, cfg, "application/geo+json, application/json")
if err != nil {
return nil, err
}
url, ok := cfg.ParamString("url", "URL")
if !ok {
return nil, fmt.Errorf("nws_forecast %q: params.url is required", cfg.Name)
}
ua, ok := cfg.ParamString("user_agent", "userAgent")
if !ok {
return nil, fmt.Errorf("nws_forecast %q: params.user_agent is required", cfg.Name)
}
return &ForecastSource{
name: cfg.Name,
url: url,
userAgent: ua,
}, nil
return &ForecastSource{http: hs}, nil
}
func (s *ForecastSource) Name() string { return s.name }
func (s *ForecastSource) Name() string { return s.http.Name }
// Kind is used for routing/policy.
func (s *ForecastSource) Kind() event.Kind { return event.Kind("forecast") }
func (s *ForecastSource) Poll(ctx context.Context) ([]event.Event, error) {
_ = ctx
return nil, fmt.Errorf("nws.ForecastSource.Poll: TODO implement (url=%s)", s.url)
raw, meta, err := s.fetchRaw(ctx)
if err != nil {
return nil, err
}
// EffectiveAt is optional; for forecasts its most naturally the run “issued” time.
// NWS gridpoint forecasts expose generatedAt (preferred) and updateTime/updated.
var effectiveAt *time.Time
switch {
case !meta.ParsedGeneratedAt.IsZero():
t := meta.ParsedGeneratedAt.UTC()
effectiveAt = &t
case !meta.ParsedUpdateTime.IsZero():
t := meta.ParsedUpdateTime.UTC()
effectiveAt = &t
}
emittedAt := time.Now().UTC()
// NWS gridpoint forecast GeoJSON commonly has a stable "id" equal to the endpoint URL.
// That is *not* unique per issued run, so we intentionally do not use it for Event.ID.
// Instead we rely on Source:EffectiveAt (or Source:EmittedAt fallback).
eventID := common.ChooseEventID("", s.http.Name, effectiveAt, emittedAt)
return common.SingleRawEvent(
s.Kind(),
s.http.Name,
standards.SchemaRawNWSHourlyForecastV1,
eventID,
emittedAt,
effectiveAt,
raw,
)
}
// ---- RAW fetch + minimal metadata decode ----
type forecastMeta struct {
// Present for GeoJSON Feature responses, but often stable (endpoint URL).
ID string `json:"id"`
Properties struct {
GeneratedAt string `json:"generatedAt"` // preferred “issued/run generated” time
UpdateTime string `json:"updateTime"` // last update time of underlying data
Updated string `json:"updated"` // deprecated alias for updateTime
} `json:"properties"`
ParsedGeneratedAt time.Time `json:"-"`
ParsedUpdateTime time.Time `json:"-"`
}
func (s *ForecastSource) fetchRaw(ctx context.Context) (json.RawMessage, forecastMeta, error) {
raw, err := s.http.FetchJSON(ctx)
if err != nil {
return nil, forecastMeta{}, err
}
var meta forecastMeta
if err := json.Unmarshal(raw, &meta); err != nil {
// If metadata decode fails, still return raw; Poll will fall back to Source:EmittedAt.
return raw, forecastMeta{}, nil
}
// generatedAt (preferred)
genStr := strings.TrimSpace(meta.Properties.GeneratedAt)
if genStr != "" {
if t, err := nwscommon.ParseTime(genStr); err == nil {
meta.ParsedGeneratedAt = t.UTC()
}
}
// updateTime, with fallback to deprecated "updated"
updStr := strings.TrimSpace(meta.Properties.UpdateTime)
if updStr == "" {
updStr = strings.TrimSpace(meta.Properties.Updated)
}
if updStr != "" {
if t, err := nwscommon.ParseTime(updStr); err == nil {
meta.ParsedUpdateTime = t.UTC()
}
}
return raw, meta, nil
}