Fix silently wrong BJD_TDB when -ov and -nea are combined - #1391
Fix silently wrong BJD_TDB when -ov and -nea are combined#1391opusbuilds wants to merge 1 commit into
Conversation
…l RA/Dec reached the time conversion unconverted With --override and --nasaexoarch together, the override wins the data choice (pDict = raw init-file values, sexagesimal strings) but the nasaexoarch branch wins the coordinate-conversion dispatch, so radec_hours_to_degree never ran. barycorrpy then raised TypeError on the string RA, and the astropy fallback parsed "19:27:06.50" with unit=deg, i.e. 19.45 degrees instead of 19h27m = 291.78 degrees, computing the barycentric correction for a point up to ~150 degrees away on the sky. Measured BJD_TDB errors: -538 s (CoRoT-2), -206 s (HAT-P-27), -93 s (TOI-1516) - at or beyond a typical Tmid error bar, roughly constant within a night, so reported mid-transit times shift by the same amount. The same unconverted string reaches the airmass code, dormant only when headers carry TELALT. Two changes: the branch dispatch now runs the conversion when both flags are set (-nea alone and -ov alone unchanged), and convert_jd_to_bjd defensively coerces string coordinates through radec_hours_to_degree so any future path delivering strings converts correctly or fails loudly instead of silently shifting the time base. The log tell for affected runs: 'barycorrpy JDUTC_to_BJDTDB conversion failed; falling back to astropy light-travel-time conversion.'
|
Re-verified against the current tip (d295d8c, 4.4.0), per the always-test-latest rule: the bug is still present there. Repro on the tip: string coordinates crash barycorrpy (TypeError), the astropy fallback engages and mis-parses hours as degrees — |
|
Re-verified on the current tip The branch cherry-picks onto |
1 participant
Fixes #1390.
Root fix: the coordinate-conversion dispatch now runs
radec_hours_to_degreewhen both flags are set (-neaalone and-ovalone unchanged). Defensive fix:convert_jd_to_bjdcoerces string coordinates before any conversion, so any future path delivering strings converts correctly or fails loudly instead of silently mis-parsing hours as degrees in the astropy fallback.Verified: string and numeric inputs now agree to 0.000000 s, barycorrpy handles the conversion (no fallback warning), and the regression test pins both the agreement and the absolute +522.5 s offset for the CoRoT-2 geometry so a sign error can't return. Full measurement table and cause chain in #1390. Test suites: only the two pre-existing
*windows*failures, unrelated (platform-specific, run on Linux).