Application-Level Forward Erasure Correction (AL-FEC) codes are a key element of telecommunication systems. They are used to recover from packet erasures during large scale content distribution, for instance within the FLUTE/ALC (file transfers) and FECFRAME (continuous real-time media transfers) protocols of 3GPP Multimedia Broadcast and Multicast Services (MBMS). However currently standardized and deployed AL-FEC codes for these protocols (e.g., Raptor(Q) or LDPC-Staircase) are all linear block codes that require the data flow to be segmented into blocks of predefined size. Surprisingly convolutional codes that are based on a sliding encoding window have not yet been considered in spite of their benefits in terms of reduced decoding latency. This work analyzes both types of codes in the context of real-time flows. More precisely, it details how to initialize block and convolutional AL-FEC codes to comply with real-time constraints and introduces the "decoding beyond maximum latency" optimization to convolutional codes. Then it compares the erasure recovery performance, the added FEC-related latency, and the decoding throughput of the two codecs. This work highlights the major benefits of convolutional codes for the large scale distribution of real-time flows from all the view points and supports the idea of extending FECFRAME specifications (RFC 6363) to support convolutional FEC codes as well.