Error paths

Situation	Behaviour
Not a single row, or no instrument able to produce a report	NULL
An instrument has no valid point left after conversion	that instrument is left out of the result
A benchmark symbol has no row in the table	Error no row for the benchmark symbol '…'
The benchmark list disagrees between instruments (entries or order)	Error every symbol must use the same benchmark list
The benchmark list holds an empty string / a NULL element / a duplicate	Error must not contain an empty string / … a NULL element / lists '…' twice
benchmark_title entry missing / empty / NULL / extra	Not an error: a missing entry falls back to that benchmark's symbol, extra entries are ignored
Price branch: the benchmark yields no return (fewer than two valid points)	Error produced no returns
periods_per_year = 0	Error periods_per_year must be greater than 0
output_dir = ''	Error output_dir must not be an empty string
The output_dir contains a NUL byte	Error contains a NUL byte
The output_dir cannot be written (missing directory, unwritable path, …)	Error cannot write the report to '<path>': … naming the path
open_in_browser with an output_dir that is not a local path (s3://…, memory://…)	Error only local file paths can be opened in a browser
output_dir in a wasm build (the browser blocks on this site)	Not an error: nothing is written and file_path is NULL
lang with no entries in the translation table	Error no translations for language '…', pointing at qs_list_translations()
qs_set_translation with a key that is not in the catalog	Error unknown key '…'
qs_set_translation with an empty language tag	Error language must not be an empty string
qs_set_translation with a NULL element in the list, a missing key, or the same key twice	Error must not contain a NULL element / every entry needs a 'key' / key '…' appears twice in one call
qs_set_translation deleting a key or a language that is not there	Not an error: it returns false (the table did not change)

Every error message starts with the registered function name (qs_html_reports: …), so it is obvious which function reported it.

Which failure surfaces as what:

The three worth seeing

Each block below is meant to fail: click Run and the message appears where the result would have been.

A benchmark that is not in the table — benchmark names symbols, and each one has to have rows:

WITH prices AS (
    SELECT * FROM read_csv('https://shijianjs.github.io/duckfn-quantstats/demo/prices.csv')
)
SELECT unnest(qs_html_reports_by_prices(
           symbol, date, price,
           {'benchmark': ['NDX']}::qs_html_report_options)) AS report
FROM prices;

Configuration mistakes are caught before anything is rendered, so they cost nothing:

WITH prices AS (
    SELECT * FROM read_csv('https://shijianjs.github.io/duckfn-quantstats/demo/prices.csv')
)
SELECT unnest(qs_html_reports_by_prices(
           symbol, date, price,
           {'periods_per_year': 0}::qs_html_report_options)) AS report
FROM prices;

An empty string in the benchmark list is a mistake that cannot be guessed around — an empty symbol can be neither a report's label nor a file name:

WITH prices AS (
    SELECT * FROM read_csv('https://shijianjs.github.io/duckfn-quantstats/demo/prices.csv')
)
SELECT unnest(qs_html_reports_by_prices(
           symbol, date, price,
           {'benchmark': ['SPX', '']}::qs_html_report_options)) AS report
FROM prices;