Read the column names from the first existing step 3 result file - #143

Problem

_get_column_names reads the column names of the step 3 result files from analysis/channelN/1_led_positions.csv, while create_binary_data leaves images without results out of the binary. An image that could not be read has no result file at all, so if image 1 is the unreadable one, --analysis stops before reaching the loop that would leave it out:

FileNotFoundError: analysis/channel0/1_led_positions.csv not found.

Reported by Manuel Osburg in #131, who hit it while checking the interplay with #137. It is not specific to corrupted images: any channel whose lowest analysed image ID is not 1 runs into it.

Change

A new _find_first_result_file collects the image IDs of the {id}_led_positions.csv files of a channel, compares them as numbers (so 2 comes before 10) and returns the path of the lowest one. _get_column_names reads the names from that file. A channel without any result file now raises a FileNotFoundError naming the directory and saying to run the data extraction first, instead of numpy's bare 1_led_positions.csv not found.

The regex that matches the result files is duplicated from _find_analysed_img_ids in step_3_functions, which cannot be imported here because it imports read_table from this module.

Create And Fill Config Analysis gained a num_ref_images argument, which defaults to the 1 it used so far.

Not covered: image 1 is also the default reference image. With num_ref_images = 1 and no results for image 1 the reference intensities come back empty, which this change does not address — whether a missing reference image should fall back or fail loudly is a separate question.

Verification

• Acceptance suite: 18/18 pass, including a new case in 07_test_corrupted_images.robot that truncates test_img_1.jpg, runs -s3_fast and --analysis, and asserts that the coefficients leave image 1 out while every remaining image keeps its own experiment time. It takes the reference from image 2, for the reason above.
• Unit tests: 25/25 pass, six of them new in ledsa/tests/UnitTests/test_file_handling.py: numeric ordering of the IDs, a missing image 1, other files in the directory ignored, an empty channel, a missing channel directory, and _get_column_names on a channel whose first result file is image 2.
• Checked the failure before the change: a channel holding only 2_led_positions.csv raises the FileNotFoundError quoted above, and returns the column names with the change.

🤖 Generated with Claude Code