Allophones are defined by means of complementary distribution. As I understand it, a complementary distribution is a "mutually exclusive" relationship between two phones, with regard to a certain phonetic environment. That is, one of the phones will only be found in that environment (and nowhere else) while the other phone will never be found in that environment (but will be found somewhere else).
For example, In English, [pʰ] and [p] are allophones of the phoneme /p/, since [pʰ] can be found at the beginning of syllables ([pʰɪn]) and nowhere else. Likewise, [p] is never found at the beginning of syllables, but can be found in other positions ([spɪn]). Summarizing in a table:
Now, in Brazilian Portuguese, [w] and [l] are considered allophones of /l/. The reasoning is similar: [l] is never found at the end of syllables, but can be found in other places ([a.'kli.vi], "slope"). [w], on the other hand, is found at the end of syllables ([saw], "salt"). The problem is: [w] can also be found elsewhere ([a.'kwi;.fe.ru], "aquifer"). The table for this distribution, therefore, would be like this:
This means that, in some environments both [w] and [l] can be found, as in the [a.'kli.vi] versus [a'.kwi.fe.ru] example. So, evidently, I'm failing to understand what "complementary distribution" really is. Any help to clarify the concept would be much appreciated!