The truncation problem of input senquence

Hi all,

Noticed the problem that if we truncate the abstraction to 1000 tokens would loss the buggy code, for example, two Defects4J bugs  closure_65 and closure_73 without  <start_bug> and <end_bug> identifiers in the generated input sequences after the truncation.  