…data

…ression-tests=

Regression tests usually don't have many or big objects, so they shouldn't take a lot of memory. If they do, the query is likely not optimal.

If the value (10 MB) does not work we can also tweak it later.

This can also be used by special tests that have a large "payload" field to ensure this field never makes it into query memory.

Needed to refactor one query because it was one query that performed a lot of filters.

The original code incorrectly assumed that a log4js logger had its own level. Instead, changing a logger's level changes the level of the logger's category system-wide. That leads to very confusing behavior that depends on the order in which you create projects or database adapters.

We're still changing global state for the trace level, but in a direct and more understandable way.

This was not buggy, but it's also unnecessarily complicated in these situations.

The two ways introduce unnecessary complexity. We also will add a feature to output AQL strings for each operation, which requires operation names. Also, some of the anonymous operations were better split into multiple named operations.

There are quite a lot of files possible per test, so this is easier to scan.

Also, we will add AQL files where multiple files will be generated per test (one per operation).

Also, it's now easier to copy a test by just copying the directory.

access-restrictions/logistics-reader was accidentally deleted in eed1127, restored.

This has mainly two use cases

- Getting a general feeling of how queries are translated into AQL to discover potential for optimizations and simplifications
- Double-checking that a code change does not change AQL in a bad way, even if you don't notice it in the result with the test data

…s-expected

The proper formatting is done by prettier via husky, but this part is easy to fix

This is the recommended value for Node 18: https://github.com/tsconfig/bases/blob/main/bases/node18.json

It includes findLastIndex() which is useful for us

This avoids variable names changing when features like sorting and filtering are added or removed, causing changes in .aql files of the regression tests

The null checks prevent arangodb from applying the reduce-extraction-to-projection optimization.

… in ArangoDB through dangling-edge filter

We currently do not trust edges - if the document they point to does not exist, we ignore the edge.

In the past, this access was done via $node != null. This has the problem that ArangoDB considers it a full document access and no longer applies the reduce-extraction-to-projection optimization rule. This rule is very important to limit query memory limit usage.

Removing the dangling edges filter completely would be breaking, but we can optimize them so the reduce-extraction-to-projection optimization still works.

Instead of collecting { root, obj } pairs in a list, we now generate the whole collect result including field selection in one subquery.

WIP: InMemoryAdapter (do this last when we're sure about the query tree API)