|
| 1 | +# patterns-json-safety |
| 2 | + |
| 3 | +JSON unmarshaling pitfalls and type flexibility patterns. |
| 4 | + |
| 5 | +--- |
| 6 | + |
| 7 | +## The Problem (8+ PRs) |
| 8 | + |
| 9 | +APIs are inconsistent about types. The same field might be: |
| 10 | +- String in one endpoint: `{"id": "12345"}` |
| 11 | +- Number in another: `{"id": 12345}` |
| 12 | +- Sometimes null: `{"id": null}` |
| 13 | + |
| 14 | +Go's strict typing causes runtime errors: |
| 15 | + |
| 16 | +``` |
| 17 | +json: cannot unmarshal number into Go struct field Group.id of type string |
| 18 | +``` |
| 19 | + |
| 20 | +--- |
| 21 | + |
| 22 | +## JSON Type Mismatch Patterns |
| 23 | + |
| 24 | +### ID Fields (Most Common) |
| 25 | + |
| 26 | +**WRONG - assumes string:** |
| 27 | +```go |
| 28 | +type Group struct { |
| 29 | + ID string `json:"id"` // Fails if API returns 12345 (number) |
| 30 | +} |
| 31 | +``` |
| 32 | + |
| 33 | +**CORRECT - use json.Number:** |
| 34 | +```go |
| 35 | +type Group struct { |
| 36 | + ID json.Number `json:"id"` // Handles both "12345" and 12345 |
| 37 | +} |
| 38 | + |
| 39 | +// Usage |
| 40 | +groupID := group.ID.String() // Always get string |
| 41 | +``` |
| 42 | + |
| 43 | +### Boolean Fields |
| 44 | + |
| 45 | +**WRONG - assumes bool:** |
| 46 | +```go |
| 47 | +type User struct { |
| 48 | + Active bool `json:"active"` // Fails if API returns "true" or 1 |
| 49 | +} |
| 50 | +``` |
| 51 | + |
| 52 | +**CORRECT - flexible bool:** |
| 53 | +```go |
| 54 | +type User struct { |
| 55 | + Active FlexibleBool `json:"active"` |
| 56 | +} |
| 57 | + |
| 58 | +type FlexibleBool bool |
| 59 | + |
| 60 | +func (f *FlexibleBool) UnmarshalJSON(data []byte) error { |
| 61 | + // Handle: true, false, "true", "false", 1, 0 |
| 62 | + var b bool |
| 63 | + if err := json.Unmarshal(data, &b); err == nil { |
| 64 | + *f = FlexibleBool(b) |
| 65 | + return nil |
| 66 | + } |
| 67 | + var s string |
| 68 | + if err := json.Unmarshal(data, &s); err == nil { |
| 69 | + *f = FlexibleBool(s == "true" || s == "1") |
| 70 | + return nil |
| 71 | + } |
| 72 | + var n int |
| 73 | + if err := json.Unmarshal(data, &n); err == nil { |
| 74 | + *f = FlexibleBool(n != 0) |
| 75 | + return nil |
| 76 | + } |
| 77 | + return fmt.Errorf("cannot unmarshal %s into bool", data) |
| 78 | +} |
| 79 | +``` |
| 80 | + |
| 81 | +### Nullable Fields |
| 82 | + |
| 83 | +**WRONG - no null handling:** |
| 84 | +```go |
| 85 | +type User struct { |
| 86 | + Email string `json:"email"` // Fails if API returns null |
| 87 | +} |
| 88 | +``` |
| 89 | + |
| 90 | +**CORRECT - use pointer:** |
| 91 | +```go |
| 92 | +type User struct { |
| 93 | + Email *string `json:"email"` // nil for null |
| 94 | +} |
| 95 | + |
| 96 | +// Usage |
| 97 | +var email string |
| 98 | +if user.Email != nil { |
| 99 | + email = *user.Email |
| 100 | +} |
| 101 | +``` |
| 102 | + |
| 103 | +--- |
| 104 | + |
| 105 | +## FlexibleID Pattern |
| 106 | + |
| 107 | +For IDs that might be string or number: |
| 108 | + |
| 109 | +```go |
| 110 | +type FlexibleID string |
| 111 | + |
| 112 | +func (f *FlexibleID) UnmarshalJSON(data []byte) error { |
| 113 | + // Try string first (most common) |
| 114 | + var s string |
| 115 | + if err := json.Unmarshal(data, &s); err == nil { |
| 116 | + *f = FlexibleID(s) |
| 117 | + return nil |
| 118 | + } |
| 119 | + |
| 120 | + // Try number |
| 121 | + var n json.Number |
| 122 | + if err := json.Unmarshal(data, &n); err == nil { |
| 123 | + *f = FlexibleID(n.String()) |
| 124 | + return nil |
| 125 | + } |
| 126 | + |
| 127 | + return fmt.Errorf("id must be string or number, got: %s", data) |
| 128 | +} |
| 129 | + |
| 130 | +func (f FlexibleID) String() string { |
| 131 | + return string(f) |
| 132 | +} |
| 133 | +``` |
| 134 | + |
| 135 | +**Usage:** |
| 136 | +```go |
| 137 | +type Group struct { |
| 138 | + ID FlexibleID `json:"id"` |
| 139 | + Name string `json:"name"` |
| 140 | +} |
| 141 | + |
| 142 | +// Works for both: |
| 143 | +// {"id": "abc123", "name": "Admins"} |
| 144 | +// {"id": 12345, "name": "Admins"} |
| 145 | + |
| 146 | +resourceID := group.ID.String() // Always string |
| 147 | +``` |
| 148 | + |
| 149 | +--- |
| 150 | + |
| 151 | +## API Response Variations |
| 152 | + |
| 153 | +Watch for these API inconsistencies: |
| 154 | + |
| 155 | +| Field | Variation 1 | Variation 2 | Solution | |
| 156 | +|-------|-------------|-------------|----------| |
| 157 | +| ID | `"12345"` | `12345` | `json.Number` or `FlexibleID` | |
| 158 | +| Active | `true` | `"true"` or `1` | `FlexibleBool` | |
| 159 | +| Count | `0` | `null` | `*int` | |
| 160 | +| Email | `"a@b.com"` | `null` | `*string` | |
| 161 | +| Timestamp | `"2024-01-01"` | `1704067200` | Custom unmarshaler | |
| 162 | + |
| 163 | +--- |
| 164 | + |
| 165 | +## Empty vs Null vs Missing |
| 166 | + |
| 167 | +Different meanings, different handling: |
| 168 | + |
| 169 | +```go |
| 170 | +type User struct { |
| 171 | + // Missing key and null both become nil |
| 172 | + Email *string `json:"email,omitempty"` |
| 173 | + |
| 174 | + // Distinguish missing from null |
| 175 | + Name NullableString `json:"name"` |
| 176 | +} |
| 177 | + |
| 178 | +type NullableString struct { |
| 179 | + Value string |
| 180 | + IsNull bool |
| 181 | + Present bool |
| 182 | +} |
| 183 | + |
| 184 | +func (n *NullableString) UnmarshalJSON(data []byte) error { |
| 185 | + n.Present = true |
| 186 | + if string(data) == "null" { |
| 187 | + n.IsNull = true |
| 188 | + return nil |
| 189 | + } |
| 190 | + return json.Unmarshal(data, &n.Value) |
| 191 | +} |
| 192 | +``` |
| 193 | + |
| 194 | +--- |
| 195 | + |
| 196 | +## Array vs Single Object |
| 197 | + |
| 198 | +Some APIs return single item as object, multiple as array: |
| 199 | + |
| 200 | +```go |
| 201 | +// API might return: |
| 202 | +// {"users": {"id": "1"}} - single user |
| 203 | +// {"users": [{"id": "1"}, ...]} - multiple users |
| 204 | + |
| 205 | +type Response struct { |
| 206 | + Users FlexibleArray[User] `json:"users"` |
| 207 | +} |
| 208 | + |
| 209 | +type FlexibleArray[T any] []T |
| 210 | + |
| 211 | +func (f *FlexibleArray[T]) UnmarshalJSON(data []byte) error { |
| 212 | + // Try array first |
| 213 | + var arr []T |
| 214 | + if err := json.Unmarshal(data, &arr); err == nil { |
| 215 | + *f = arr |
| 216 | + return nil |
| 217 | + } |
| 218 | + |
| 219 | + // Try single object |
| 220 | + var single T |
| 221 | + if err := json.Unmarshal(data, &single); err == nil { |
| 222 | + *f = []T{single} |
| 223 | + return nil |
| 224 | + } |
| 225 | + |
| 226 | + return fmt.Errorf("expected array or object") |
| 227 | +} |
| 228 | +``` |
| 229 | + |
| 230 | +--- |
| 231 | + |
| 232 | +## Detection in Code Review |
| 233 | + |
| 234 | +**Red flags:** |
| 235 | +1. `ID string` for API response structs - should be `json.Number` or `FlexibleID` |
| 236 | +2. `bool` for API fields without checking API consistency |
| 237 | +3. Non-pointer types for optional fields |
| 238 | +4. No custom unmarshalers for inconsistent APIs |
| 239 | + |
| 240 | +**Questions to ask:** |
| 241 | +- "What types does this API actually return? Did you check multiple endpoints?" |
| 242 | +- "What happens if this field is null?" |
| 243 | +- "Does the API always return this as a string, or sometimes a number?" |
| 244 | + |
| 245 | +--- |
| 246 | + |
| 247 | +## Testing JSON Handling |
| 248 | + |
| 249 | +```go |
| 250 | +func TestFlexibleID_Unmarshal(t *testing.T) { |
| 251 | + tests := []struct { |
| 252 | + input string |
| 253 | + expected string |
| 254 | + }{ |
| 255 | + {`{"id": "abc123"}`, "abc123"}, |
| 256 | + {`{"id": 12345}`, "12345"}, |
| 257 | + {`{"id": 0}`, "0"}, |
| 258 | + } |
| 259 | + |
| 260 | + for _, tt := range tests { |
| 261 | + var obj struct { |
| 262 | + ID FlexibleID `json:"id"` |
| 263 | + } |
| 264 | + err := json.Unmarshal([]byte(tt.input), &obj) |
| 265 | + if err != nil { |
| 266 | + t.Errorf("failed to unmarshal %s: %v", tt.input, err) |
| 267 | + } |
| 268 | + if obj.ID.String() != tt.expected { |
| 269 | + t.Errorf("got %s, want %s", obj.ID, tt.expected) |
| 270 | + } |
| 271 | + } |
| 272 | +} |
| 273 | +``` |
0 commit comments