Search function in synthesis information block not working correctly

[KubaLemanski]

There is an issue with the search function in the synthesis information block. When searching for compounds they seem to appear only when certain parts of the name are typed in. For example, when searching for ‘vanadium(II) chloride’ it will not appear in the list when either ‘v’ or ‘vanadium(II)’ is searched, but will appear when ‘vanadium’ is searched. I also noticed this when searching for samples created within datalab, not just those from the inventory. For some compounds this is ok, but some won’t appear seemingly regardless of what the search input is.

[ml-evs]

I would argue it's working correctly but not well 🙃

I've just replicated your case -- completely agree it should work. The problem is that what you type into that box is actually a regular expression, and brackets have a special meaning in them that could allow for complex queries. If you try searching for vanadium\(II\) (i.e., escaping the brackets) or just vanadium\(, then your search should work. We should maybe do that escaping for you by default, but will then have to have another hidden way to escape brackets so that you can use a real regexp...

This is the same issue when searching for just "v"; its not looking necessarily for the words that start with v, but just words containing v (or V). I imagine in your case, pretty much every sample has a word with "v" in it, so its flooding your results. Again, perhaps by default, we should match only from the start of words...

Hopefully this gives you a workaround, but this is something we should try to fix by adding some simple heuristics to the user input (e.g., restrict the user search to apply to the start of strings only, and don't let them use brackets).

[ml-evs]

Just messed around with this in #1338 and think I have some reasonable fixes (though the db queries themselves are now much more expensive).

If a sample has vanadium(II) oxide somewhere in it, the following should now return better results:

• ✔️ v -- by adding an implicit word boundary at the start of the regex
• ✔️ vanadium(II) -- by treating brackets as literal chars
• ✔️ vanadium oxide -- by splitting the user query by spaces and chaining them as MATCH \bvanadium AND \boxide
• ❌ xide or adium -- no longer matches from word boundary

If this sounds reasonable @KubaLemanski, I can deploy this as a patch on @MJCliffe's server

[ml-evs]

In fact, @KubaLemanski & @MJCliffe, I back myself and reckon this works quite well, have deployed on your server now. Let me know if this somehow makes things worse and I can roll back, but I think this does improve things at the expense of speed.

Please give some feedback here -- will close the issue once you're happy. Thanks a lot for reporting it!

[ml-evs]

We can maybe tweak it to not only match word boundaries but also things like *bismuthate* in the middle of a word given some of the long compound names in your database, perhaps some heuristic for using word boundaries if the query is short, e.g. "van" should match "vanadium" but not "caravan" or "galvanyse", but "ravan" matches "caravan" and "caravanning"

[ml-evs]

Have also just deployed #1338 to the demo deployment if @jdbocarsly wanted to play around with it