I don't expect perfection, I'm talking about a measure of usefulness. Copilot's code needs to be brought in to project standards (the extent of which depends on language - Java code and PHP code are a measure of the large quantity of legacy that Copilot had access to).
TabNine for example does better at completing code in local context (evening it can't write complete functions) because it learns from local context. It will never autocomplete non existent variables for example.
That's a bit of trade-off obviously and largely depends on language.
I've been using Copilot with both Go and PHP and ended up trusting Go autocompletes (with some variable names fixing) but refactoring PHP code in 90% of cases (eg: it would write functions with PHP 7 code in situations where TabNine could tell I was using PHP 8)
I don't expect perfection, I'm talking about a measure of usefulness. Copilot's code needs to be brought in to project standards (the extent of which depends on language - Java code and PHP code are a measure of the large quantity of legacy that Copilot had access to).
TabNine for example does better at completing code in local context (evening it can't write complete functions) because it learns from local context. It will never autocomplete non existent variables for example.
That's a bit of trade-off obviously and largely depends on language.
I've been using Copilot with both Go and PHP and ended up trusting Go autocompletes (with some variable names fixing) but refactoring PHP code in 90% of cases (eg: it would write functions with PHP 7 code in situations where TabNine could tell I was using PHP 8)
Considered, maybe we will see improvements in its upcoming commercial version.