read_file should indicate when the model tries to read out of bounds #274120
Description
The model tried to read from a line number that doesn't exist twice:
The file only had 498 lines. Here is the first tool call:
# Tool Call - toolu_01EPA3tz4EK5HHFca39CDWfh__vscode-1761817999475
## Request
~~~
id : toolu_01EPA3tz4EK5HHFca39CDWfh__vscode-1761817999475
tool : read_file
args : {
"filePath": "/Users/alex/src/vscode/src/vs/editor/test/common/model/tokensStore.test.ts",
"offset": 535,
"limit": 50
}
~~~
## Response
~~~
```typescript
```
~~~
Notice how we don't communicate to the model that this file doesn't have 535 lines at all. Then the model goes and tries again:
# Tool Call - toolu_01XFdtzdCYw5ViMycduyhJon__vscode-1761817999476
## Request
~~~
id : toolu_01XFdtzdCYw5ViMycduyhJon__vscode-1761817999476
tool : read_file
args : {
"filePath": "/Users/alex/src/vscode/src/vs/editor/test/common/model/tokensStore.test.ts",
"limit": 20,
"offset": 525
}
~~~
## Response
~~~
```typescript
```
~~~
The model is just trying to guess its way to the file size. IMHO we should report an error when the offset is > the line count in the file and explain in the error the actual file's line count. This should help the model recover faster and avoid doing two invalid read file tool calls in a row.
