Different standard errors in reghdfe 3.2.7 versus reghdfe 5.7.3

[liaoguanmin]

Hello Sergio,
I am running a specification with reghdfe. For the same regression in which standard errors are clustered at both region and year levels. However, reghdfe 3.2.7 and reghdfe 5.7.3 give different standard errors. If I suppress the constant term in reghdfe 5.7.3, the difference becomes smaller but it still exists. Could you please tell me the reason for this? Thank you very much!
Best regards,
Guanmin

[ozak]

Hi @sergiocorreia and @liaoguanmin.
I was replicating some earlier results in one of my projects, and I just noticed a similar issue between reghdfe-5.6.8 and reghdfe-6.12.3. With reghdfe-5.6.8 the regression produces multiway clustered SEs, while reghdfe-6.12.3 fails to compute any SEs = ( . ) . I was looking at the changes in the Recent Updates and I do not see any mention of bugs or any reference to SEs issues. Do you have any idea what may have changed that may cause this?