fix(ai-sdk): model provider attribute transformation for openai

[galkleinman]

Important

Refactor AI SDK span transformations into ai-sdk-transformations.ts and add comprehensive tests.

• Behavior:
  • Introduces transformAiSdkSpan in ai-sdk-transformations.ts to handle AI SDK span transformations.
  • Refactors span name and attribute transformation logic from span-processor.ts to ai-sdk-transformations.ts.
  • Applies transformations in onSpanEnd in span-processor.ts using transformAiSdkSpan.
• Transformations:
  • transformAiSdkSpanName changes span names for specific AI operations.
  • transformResponseText, transformPromptMessages, transformPromptTokens, transformCompletionTokens, calculateTotalTokens, and transformVendor handle specific attribute transformations.
• Testing:
  • Adds ai-sdk-transformations.test.ts with tests for all transformation functions, ensuring correct behavior and handling of edge cases.

^{This description was created by for ad0afa6. You can customize this summary. It will automatically update as commits are pushed.}

[ellipsis-dev]

Important

Looks good to me! 👍

Reviewed everything up to 3ccdc9f in 51 seconds. Click for details.

• Reviewed 81 lines of code in 1 files
• Skipped 0 files when reviewing.
• Skipped posting 3 draft comments. View those below.
• Modify your settings and rules to customize what types of comments Ellipsis leaves. And don't forget to react with 👍 or 👎 to teach Ellipsis.

1. packages/traceloop-sdk/src/lib/tracing/span-processor.ts:177

• Draft comment:
  Delegating transformation to transformAiSdkSpan() improves modularity. Ensure that this function fully replicates all the previous inline transformations and error handling (e.g., JSON parsing in ai.prompt.messages).
• Reason this comment was not posted:
  Comment did not seem useful. Confidence is useful = 0% <= threshold 50% The comment suggests ensuring that a function replicates previous behavior, which violates the rule against asking the PR author to ensure behavior is intended or tested. It doesn't provide a specific code suggestion or ask for a specific test to be written.

2. packages/traceloop-sdk/src/lib/tracing/span-processor.ts:161

• Draft comment:
  The JSDoc comment still mentions 'Vercel AI compatibility'. Update it to reflect the more general AI SDK span transformations now applied.
• Reason this comment was not posted:
  Comment was not on a location in the diff, so it can't be submitted as a review comment.

3. packages/traceloop-sdk/src/lib/tracing/span-processor.ts:170

• Draft comment:
  Legacy inline transformation logic for Vercel AI Adapters was removed. Confirm that transformAiSdkSpan now handles all attribute mappings (e.g., span name remapping, response text and prompt message adjustments, token usage calculations) as before.
• Reason this comment was not posted:
  Comment was not on a location in the diff, so it can't be submitted as a review comment.

Workflow ID: wflow_wbhErfoXRRs3Nvg1

^{You can customize by changing your verbosity settings, reacting with 👍 or 👎, replying to comments, or adding code review rules.}

[ellipsis-dev]

Caution

Changes requested ❌

Reviewed 40c6ea4 in 1 minute and 47 seconds. Click for details.

• Reviewed 662 lines of code in 2 files
• Skipped 0 files when reviewing.
• Skipped posting 2 draft comments. View those below.
• Modify your settings and rules to customize what types of comments Ellipsis leaves. And don't forget to react with 👍 or 👎 to teach Ellipsis.

1. packages/traceloop-sdk/src/lib/tracing/ai-sdk-transformations.ts:48

• Draft comment:
  Suggestion: In transformPromptMessages, the catch block silently swallows JSON parsing errors. Consider logging the error for easier debugging.
• Reason this comment was not posted:
  Decided after close inspection that this draft comment was likely wrong and/or not actionable: usefulness confidence = 20% vs. threshold = 50% This is a data transformation layer where parsing errors likely indicate malformed input data. The current approach of silently skipping seems intentional - it allows the transformation to continue with other attributes even if one fails. Adding logging could create noise in production logs since this isn't necessarily an error condition. The comment is about code quality but may not be a clear improvement. The suggestion could help with debugging issues in development. Silent failures can make issues harder to track down. However, this appears to be an intentional design choice to gracefully handle malformed data without disrupting the flow. Logging every parse failure could create unnecessary noise for what is essentially a non-critical transformation step. The current silent failure approach appears to be an intentional design choice for graceful degradation. Adding logging would be a matter of preference rather than a clear improvement.

2. packages/traceloop-sdk/test/ai-sdk-transformations.test.ts:339

• Draft comment:
  Testing Improvement: Consider adding tests for calculateTotalTokens with token values of 0 to ensure the sum is computed correctly.
• Reason this comment was not posted:
  Decided after close inspection that this draft comment was likely wrong and/or not actionable: usefulness confidence = 0% vs. threshold = 50% While testing zero values is generally good practice, in this case the function is just doing basic addition. The existing tests already cover type conversion and edge cases. Zero is not a special case for addition - if the function works for 50+25, it will work for 0+0. The function is simple enough that zero values don't introduce any special behavior. I could be underestimating the importance of explicitly testing boundary values like zero. There could be type coercion issues specific to zero in JavaScript. The existing test with string values already verifies type coercion behavior. Zero addition is a fundamental operation that doesn't need special testing beyond normal values. The comment should be deleted as the suggested test would be redundant given the existing comprehensive test coverage.

Workflow ID: wflow_q5nfLC0RIbBWwHwv

^{You can customize by changing your verbosity settings, reacting with 👍 or 👎, replying to comments, or adding code review rules.}

[ellipsis-dev]

Bug: The condition in calculateTotalTokens uses 'if (promptTokens && completionTokens)' which fails when token values are 0 (a valid case). Consider checking for undefined (e.g., promptTokens != null && completionTokens != null) instead.

Suggested change

	if (promptTokens && completionTokens) {
	if (promptTokens != null && completionTokens != null) {

[ellipsis-dev]

Important

Looks good to me! 👍

Reviewed ad0afa6 in 1 minute and 6 seconds. Click for details.

• Reviewed 661 lines of code in 2 files
• Skipped 0 files when reviewing.
• Skipped posting 2 draft comments. View those below.
• Modify your settings and rules to customize what types of comments Ellipsis leaves. And don't forget to react with 👍 or 👎 to teach Ellipsis.

1. packages/traceloop-sdk/src/lib/tracing/ai-sdk-transformations.ts:81

• Draft comment:
  Potential issue: using a truthy check for promptTokens and completionTokens may skip valid 0 values. Consider checking for undefined instead, e.g. (promptTokens !== undefined && completionTokens !== undefined).
• Reason this comment was not posted:
  Comment was on unchanged code.

2. packages/traceloop-sdk/src/lib/tracing/ai-sdk-transformations.ts:88

• Draft comment:
  Consider validating that the 'vendor' is a string before calling startsWith to avoid runtime errors if an unexpected type is provided.
• Reason this comment was not posted:
  Comment was not on a location in the diff, so it can't be submitted as a review comment.

Workflow ID: wflow_LIWhVBhgT7VeBXTi

^{You can customize by changing your verbosity settings, reacting with 👍 or 👎, replying to comments, or adding code review rules.}