+2-pass DER encoding
+-------------------
+
+There is ability to do 2-pass encoding to DER, writing results directly
+to specified writer (buffer, file, whatever). It could be 1.5+ times
+slower than ordinary encoding, but it takes little memory for 1st pass
+state storing. For example, 1st pass state for CACert.org's CRL with
+~416K of certificate entries takes nearly 3.5 MB of memory.
+``SignedData`` with several gigabyte ``EncapsulatedContentInfo`` takes
+nearly 0.5 KB of memory.
+
+If you use :ref:`mmap-ed <mmap>` memoryviews, :ref:`SEQUENCE OF
+iterators <seqof-iterators>` and write directly to opened file, then
+there is very small memory footprint.
+
+1st pass traverses through all the objects of the structure and returns
+the size of DER encoded structure, together with 1st pass state object.
+That state contains precalculated lengths for various objects inside the
+structure.
+
+::
+
+ fulllen, state = obj.encode1st()
+
+2nd pass takes the writer and 1st pass state. It traverses through all
+the objects again, but writes their encoded representation to the writer.
+
+::
+
+ opener = io.open if PY2 else open
+ with opener("result", "wb") as fd:
+ obj.encode2nd(fd.write, iter(state))
+
+.. warning::
+
+ You **MUST NOT** use 1st pass state if anything is changed in the
+ objects. It is intended to be used immediately after 1st pass is
+ done!
+
+If you use :ref:`SEQUENCE OF iterators <seqof-iterators>`, then you
+have to reinitialize the values after the 1st pass. And you **have to**
+be sure that the iterator gives exactly the same values as previously.
+Yes, you have to run your iterator twice -- because this is two pass
+encoding mode.
+
+If you want to encode to the memory, then you can use convenient
+:py:func:`pyderasn.encode2pass` helper.
+